{"id":226,"date":"2026-09-04T11:14:02","date_gmt":"2026-09-04T03:14:02","guid":{"rendered":"http:\/\/47.236.250.150\/?p=226"},"modified":"2026-09-04T20:59:58","modified_gmt":"2026-09-04T12:59:58","slug":"qwen3-8-max-a-new-bar-for-coding-and-cowork","status":"publish","type":"post","link":"https:\/\/tongyi.ai\/?p=226","title":{"rendered":"Qwen3.8-Max: A New Bar for Coding and Cowork"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">Today, we are officially releasing&nbsp;<strong>Qwen 3.8-Max<\/strong>, the most capable model in the Qwen family to date. This also marks the first time we will open-source the weights of a Qwen-Max-class model \u2014 the open weights will be released next week. Built upon the architectural foundation of Qwen 3.5, Qwen 3.8-Max scales to&nbsp;<strong>2.4 trillion<\/strong>&nbsp;parameters, delivering comprehensive improvements across coding, work, research, and long-horizon tasks. It can not only answer more challenging questions, but also complete complex tasks end-to-end with greater reliability, producing dependable deliverables.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Qwen3.8-Max<\/strong>&nbsp;\u2014 now available via&nbsp;<a href=\"https:\/\/www.qwencloud.com\/\" target=\"_blank\" rel=\"noopener\">QwenCloud<\/a>:\n<ul class=\"wp-block-list\">\n<li>2.4T parameters (95B active), with open weights releasing next week<\/li>\n\n\n\n<li>comprehensive improvements across coding, work, research, and long-horizon tasks<\/li>\n\n\n\n<li>end-to-end and dependable delivery of complex tasks<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>Call via API on&nbsp;<a href=\"https:\/\/www.qwencloud.com\/\" target=\"_blank\" rel=\"noopener\">QwenCloud<\/a>.<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/qianwen-res.oss-cn-beijing.aliyuncs.com\/Qwen3.8\/performance.png\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">Coding<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">For a top model, coding today means far more than writing a function on request \u2014 it means taking a real, multi-day project from an empty folder all the way to a finished result, on its own. We tested Qwen3.8-Max on three such challenges, where every result had to be earned by actually writing and running code, with&nbsp;<strong>no human help at all<\/strong>. One thread runs through all three: Qwen3.8-Max doesn\u2019t just follow a fixed plan \u2014 it&nbsp;<strong>self-evolves through feedback loops<\/strong>, whether that means building a harness that upgrades itself, refining a research method experiment after experiment, or climbing a competition leaderboard submission after submission.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">10+ Days of Autonomous Coding: Building a Self-Evolving Harness<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">In this case, Qwen3.8-Max was asked to create the&nbsp;<code>oh-my-cli<\/code>&nbsp;project from scratch and, over a 10+ day long-horizon autonomous coding run, build a self-evolving harness. It brings user feedback, advanced community practices, and the model\u2019s own self-test results into one engineering loop: requirements are normalized into issues, automatically claimed and executed by agents, and continuously iterated through code, tests, previews, and logs. The complete project trace is publicly available in the GitHub repository&nbsp;<a href=\"https:\/\/github.com\/qwen-code-dev-bot\/oh-my-cli\">qwen-code-dev-bot\/oh-my-cli<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Key implementation details in the autonomous coding harness:<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Loop Engineering Setup: task state, dispatch, and recovery.<\/strong>&nbsp;Qwen3.8-Max combines an issue state machine, dispatcher, monitor, and watchdog into one execution loop: after a new requirement enters GitHub Issues, an agent claims it through the state machine and moves through&nbsp;<code>ready \u2192 leased \u2192 active<\/code>; once implementation is complete, E2E tests and CI checks are triggered, and the PR is merged after passing.<\/li>\n\n\n\n<li><strong>Self-testing: product self-testing and maintenance.<\/strong>&nbsp;After each update, the model triggers Build, Unit Test, E2E, and Desktop Lifecycle validation; abnormal states are routed back to the relevant issue \/ PR for fixes and re-verification.<\/li>\n\n\n\n<li><strong>Multi-source Evolution:<\/strong>&nbsp;product upgrades from multiple demand signals. By converting community experience and user \/ developer feedback into executable work, the harness continuously evolves&nbsp;<code>\/goal<\/code>,&nbsp;<code>\/resume<\/code>, Dynamic Workflow, Session Replay, Desktop, and other capabilities.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">As of&nbsp;<strong>July 30, 2026<\/strong>, after approximately&nbsp;<strong>16 days<\/strong>&nbsp;of fully autonomous AI operation, the repository had accumulated&nbsp;<strong>265 commits, 127 PRs, and 151 issues<\/strong>, demonstrating a continuously evolving autonomous coding capability.<\/p>\n\n\n\n<figure class=\"wp-block-video\"><video loop poster=\"https:\/\/img.alicdn.com\/imgextra\/i4\/O1CN01a6lF5OyB6SI6UpHk_!!6000000001097-0-tps-3200-1800.jpg\" src=\"https:\/\/cloud.video.taobao.com\/vod\/1h6qXKSnAjXp6lHfSDX1xfKukA0Y5hUUQ73qqUtY4aQ.mp4\" playsinline><\/video><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Video 1. In a 10+ day long-horizon autonomous coding run, Qwen3.8-Max autonomously builds a self-evolving harness, continuously completing community requirement collection, issue dispatch, code generation, verification, and self-repair.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Reproduce a research paper \u2014 then improve it<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">We handed Qwen3.8-Max a recent research paper \u2014&nbsp;<a href=\"https:\/\/arxiv.org\/abs\/2605.22389\"><em>\u201cUnified Data Selection for LLM Reasoning\u201d<\/em><\/a>&nbsp;\u2014 and asked it to:&nbsp;<strong>reproduce the paper\u2019s experiment in code, then try to do better.<\/strong>&nbsp;The paper tackles a very practical question in AI training: when you have far more data than you can afford to train on,&nbsp;<em>which examples are actually worth keeping?<\/em>&nbsp;The paper\u2019s answer is to prize the examples full of&nbsp;<strong>\u201chard decision points\u201d<\/strong>&nbsp;\u2014 the moments in a worked solution where the model was genuinely unsure which way to go next.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The catch: Qwen3.8-Max started from&nbsp;<strong>nothing but the paper and a set of GPUs<\/strong>&nbsp;\u2014 no starter code, no ready-made pipeline. The data-processing scripts, the training code, the evaluation setup \u2014 it had to design and write&nbsp;<strong>all of it from scratch<\/strong>, exactly the kind of work that takes skilled engineers days.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Working&nbsp;<strong>completely on its own for about five days<\/strong>&nbsp;(~125 hours of continuous effort), Qwen3.8-Max wrote roughly&nbsp;<strong>7,600 lines of code<\/strong>, took over&nbsp;<strong>1,100 actions<\/strong>, and ran&nbsp;<strong>33 rounds of GPU training<\/strong>. It first spent ~37 hours rebuilding the paper\u2019s full pipeline from zero and&nbsp;<strong>reproduced its six main findings<\/strong>&nbsp;\u2014 repeatedly fine-tuning a Qwen3-8B model on the data it selected and confirming the gains on hard math benchmarks (for instance, the paper\u2019s selection method beats picking data at random by&nbsp;<strong>+7.7%<\/strong>&nbsp;on AIME24).<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Then it went further, turning reproduction into&nbsp;<strong>self-evolution<\/strong>. Over the next ~88 hours it ran a self-improving research loop \u2014&nbsp;<em>form a hypothesis \u2192 write the code \u2192 run it on GPUs \u2192 analyze \u2192 try again<\/em>&nbsp;\u2014 inventing and testing&nbsp;<strong>18 improvement ideas of its own across four rounds<\/strong>. Each round\u2019s results fed the next round\u2019s hypotheses, and by diagnosing what went wrong with each attempt it finally evolved a new method that&nbsp;<strong>beats the paper\u2019s own approach<\/strong>, a&nbsp;<strong>+2.7-point<\/strong>&nbsp;gain on the competition-level math benchmark AIME24.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Expand<\/p>\n\n\n\n<figure class=\"wp-block-embed\"><div class=\"wp-block-embed__wrapper\">\nhttps:\/\/docs.qwenlm.ai\/resources\/thKHg_ml_coding_demo_hes_reproduction_improvement.htmlHow the improvement search unfolded \u2014 4 rounds, 18 ideas\n<\/div><\/figure>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th class=\"has-text-align-left\" data-align=\"left\"><\/th><th class=\"has-text-align-left\" data-align=\"left\"><\/th><th class=\"has-text-align-left\" data-align=\"left\"><\/th><th class=\"has-text-align-left\" data-align=\"left\"><\/th><\/tr><\/thead><tbody><tr><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\"><\/td><td class=\"has-text-align-left\" data-align=\"left\"><strong><\/strong><\/td><td class=\"has-text-align-left\" data-align=\"left\"><strong><\/strong><\/td><td class=\"has-text-align-left\" data-align=\"left\"><strong><\/strong><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">Beat hundreds of human teams in 24 hours<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Next we entered Qwen3.8-Max into a real online contest \u2014 the&nbsp;<a href=\"https:\/\/tianchi.aliyun.com\/competition\/entrance\/532277\">WWW2025 Multimodal Dialogue Intent Recognition Challenge<\/a>, hosted on&nbsp;<strong>Alibaba Cloud\u2019s Tianchi platform<\/strong>, where&nbsp;<strong>526 human teams<\/strong>&nbsp;were competing. The task: read customer-service chats \u2014 both the text&nbsp;<em>and<\/em>&nbsp;the screenshots \u2014 and correctly work out what the customer wants.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Working entirely on its own and under a strict&nbsp;<strong>24-hour<\/strong>&nbsp;time limit, Qwen3.8-Max read the competition rules and built a full solution in code. For the text side, it fine-tuned and ensembled several Chinese language models \u2014&nbsp;<strong>BERT, MacBERT, and RoBERTa<\/strong>; for the product screenshots, it fine-tuned a vision-language model,&nbsp;<strong>Qwen2.5-VL-7B<\/strong>, backed by a&nbsp;<strong>Chinese-CLIP<\/strong>&nbsp;model for images its main model was unsure about. It then fused all of them into a single&nbsp;<strong>weighted-voting system<\/strong>, calibrating how much each model\u2019s vote should count through cross-validation and adding extra image voters to break ties. Across&nbsp;<strong>45 submissions<\/strong>&nbsp;\u2014 each round\u2019s feedback steering the next round of fine-tuning and re-weighting \u2014 its accuracy climbed steadily from&nbsp;<strong>0.60 to a final 0.853<\/strong>, beating&nbsp;<strong>458 of the 526 human teams (87% of the field)<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Expand<\/p>\n\n\n\n<figure class=\"wp-block-embed\"><div class=\"wp-block-embed__wrapper\">\nhttps:\/\/docs.qwenlm.ai\/resources\/QfeH4_www2025_animation_v5_1.html\n<\/div><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Together, these three cases show what makes Qwen3.8-Max stand out: it can stay focused on a hard, open-ended goal for days, come up with its own ideas, and turn them into working results \u2014 all without a human in the loop.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Work<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Alongside coding,&nbsp;<strong>real work<\/strong>&nbsp;&#8211; the messy, multi-step, tool-heavy tasks that fill the working day in nearly every profession &#8211; is the other main track where frontier models create enormous economic value. Making Qwen3.8-Max broadly competent and reliably robust across these workflows is therefore central to our mission.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Scaling Real-World RL Systems.<\/strong>&nbsp;By jointly scaling RL environments and compute, we lift&nbsp;<strong>general working competence<\/strong>&nbsp;uniformly across several popular harnesses (QwenWork \/ Claude Code \/ Codex \/ OpenClaw \/ Hermes). Achieving this required addressing three coupled challenges:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>Continuously scaling decoupled real environments<\/strong>&nbsp;along independent axes \u2014&nbsp;<em>Task<\/em>&nbsp;(single-task \u2192 multi-task \u2192 multi-day),&nbsp;<em>Workspace<\/em>&nbsp;(multi-file \u2192 hierarchical folders \u2192 complex heterogeneous folders), and&nbsp;<em>Harness<\/em>&nbsp;(category, version, skills) \u2014 so environment growth&nbsp;<strong>compounds combinatorially<\/strong>&nbsp;rather than requiring bespoke integration.<\/li>\n\n\n\n<li><strong>A Universal Reward System<\/strong>&nbsp;that internalizes heterogeneous verification \u2014 spanning execution-based checking, rubric-conditioned adjudication over text and rendered visual output, and agentic inspection \u2014 under automatically scalable rubrics. By unifying these modalities within&nbsp;<strong>one reward system<\/strong>, it provides a coherent and reliable source of reward across all environments, eliminating the inconsistency inherent in maintaining task-specific verifiers.<\/li>\n\n\n\n<li><strong>An online data balancer<\/strong>&nbsp;that shapes every batch to keep its distribution over tasks, difficulty, workspaces, and harnesses highly balanced,&nbsp;<strong>suppressing inter-batch gradient variance<\/strong>&nbsp;and thereby sustaining stable, continued scaling of RL compute.<\/li>\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\">Together these supply&nbsp;<strong>breadth, reliable reward, and stability<\/strong>&nbsp;\u2014 turning joint environment-and-compute scale into a measurable, horizontal lift in real-world working ability.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/qianwen-res.oss-cn-beijing.aliyuncs.com\/Qwen3.8\/work_showcases\/training-score-vs-envs-scale.png#center\" alt=\"Fig 1. Qwen3.8-Max shows steady, consistent gains across dozens of in-house and public working benchmarks as RL training continues to scale up.\"\/><figcaption class=\"wp-element-caption\">Fig 1. Qwen3.8-Max shows steady, consistent gains across dozens of in-house and public working benchmarks as RL training continues to scale up.<\/figcaption><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/qianwen-res.oss-cn-beijing.aliyuncs.com\/Qwen3.8\/work_showcases\/harness-generalization-3.8.png#center\" alt=\"Fig 2. Qwen3.8-Max achieves comparable performance across many harnesses, including QwenWork, Claude Code, Codex, OpenClaw, and Hermes.\"\/><figcaption class=\"wp-element-caption\">Fig 2. Qwen3.8-Max achieves comparable performance across many harnesses, including QwenWork, Claude Code, Codex, OpenClaw, and Hermes.<\/figcaption><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">Testing the&nbsp;<em>Breadth<\/em>&nbsp;of Working Ability Across&nbsp;<em>Hundreds<\/em>&nbsp;of High-Value Professions<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">As frontier models take on an ever-widening role in economically valuable work, we stress-tested the&nbsp;<strong>breadth<\/strong>&nbsp;of Qwen3.8-Max\u2019s ability to deliver production-quality results in real workflows \u2014 spanning high-frequency tasks across&nbsp;<strong>several hundred high-economic-value professions<\/strong>. A few representative showcases:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Corporate compliance counsel<\/strong>&nbsp;\u2014 Qwen3.8-Max surfaced&nbsp;<strong>1,284 relevant clauses<\/strong>&nbsp;across a corpus of&nbsp;<strong>hundreds of documents<\/strong>&nbsp;in a single pass, completing the full review in&nbsp;<strong>under an hour<\/strong>. Such a review typically takes a paralegal team working collaboratively for&nbsp;<strong>around a week<\/strong>.<\/li>\n\n\n\n<li><strong>UI\/UX designer<\/strong>&nbsp;\u2014 Qwen3.8-Max produced a high-fidelity, interactive prototype for the digital-banking app&nbsp;<em>NOVA<\/em>&nbsp;\u2014&nbsp;<strong>8 screens<\/strong>&nbsp;with a consistent design system, delivered in&nbsp;<strong>one shot<\/strong>&nbsp;with&nbsp;<strong>zero rounds<\/strong>&nbsp;of human revision, versus&nbsp;<strong>3\u20135 rounds<\/strong>&nbsp;of revision in a conventional workflow.<\/li>\n\n\n\n<li><strong>Restaurant brand founder<\/strong>&nbsp;\u2014 Qwen3.8-Max read through&nbsp;<strong>over a hundred ingredient-supply briefs<\/strong>&nbsp;and produced a complete&nbsp;<strong>26-dish menu<\/strong>&nbsp;in one pass. Each dish is annotated with its average caloric value and ingredient provenance, with the food-cost ratio held at&nbsp;<strong>33.8%<\/strong>. Such menu development would normally require a head chef and operations team weeks of iterative recipe testing, costing, and refinement.<\/li>\n\n\n\n<li><strong>Structural engineer<\/strong>&nbsp;\u2014 From a single set of drawings, Qwen3.8-Max reconstructed the seismic structural model of a 30-story office tower in the browser, with natural period, base shear, and inter-story drift ratio all available for real-time inspection on hover. In a traditional workflow, an engineer would need to build the model manually in specialized modeling software, typically taking over a week.<\/li>\n\n\n\n<li><strong>Rehabilitation therapist<\/strong>&nbsp;\u2014 Qwen3.8-Max turned a&nbsp;<strong>2D paper assessment form<\/strong>&nbsp;into a&nbsp;<strong>3D interactive demo<\/strong>&nbsp;with freely rotatable viewing angles and layer-by-layer&nbsp;<strong>anatomical overlays<\/strong>, letting patients see exactly where the injury sits and how recovery progresses \u2014 work previously outsourced to a medical-animation studio at&nbsp;<strong>2\u20134 weeks<\/strong>\u2019 lead time and&nbsp;<strong>thousands of dollars<\/strong>&nbsp;in cost.<\/li>\n\n\n\n<li><strong>Sports data analyst<\/strong>&nbsp;\u2014 Qwen3.8-Max parsed&nbsp;<strong>~8,400 offensive\/defensive possessions per player<\/strong>&nbsp;into a ready-to-use&nbsp;<strong>player tactical profile<\/strong>&nbsp;and&nbsp;<strong>coaching report<\/strong>&nbsp;in&nbsp;<strong>tens of minutes<\/strong>. A traditional analytics team would need to manually complete tactical segmentation, causal attribution, and report writing \u2014 a process typically spanning&nbsp;<strong>several working days<\/strong>.<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-video\"><video loop poster=\"https:\/\/img.alicdn.com\/imgextra\/i3\/O1CN01usJ6dI1hurI4MFFMj_!!6000000004338-0-tps-3200-1800.jpg\" src=\"https:\/\/cloud.video.taobao.com\/vod\/g1Iouht-8GIF-k6peB4RPgL0xYvGf6OK_THgInrMdb4.mp4\" playsinline><\/video><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Video 1. Across hundreds of high-value professions, Qwen3.8-Max measurably boosts human productivity in real workflows \u2014 showcasing the&nbsp;<em>breadth<\/em>&nbsp;of its working ability.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Building a Profitable End-to-End Quant Strategy in a Single Session<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Powered by its&nbsp;<strong>Dynamic Workflows<\/strong>&nbsp;construction capability, Qwen3.8-Max drives task planning programmatically and orchestrates large-scale sub-agent systems with precision \u2014 turning a single conversation into an end-to-end, automated quant-research loop.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Depth \u2014 end-to-end ETF-rotation strategy R&amp;D.<\/strong>&nbsp;From a one-line task description, Qwen3.8-Max autonomously planned a complex dynamic workflow and worked for&nbsp;<strong>hours<\/strong>&nbsp;to deliver a complete ETF-rotation strategy \u2014 building the data system, constructing base factors, and orchestrating multi-round greedy iteration, all while dynamically analyzing backtests and correcting course. Throughout, it&nbsp;<strong>acted on evidence instead of a fixed script<\/strong>:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>When it observed misalignment between design-period metrics and validation-period metrics \u2014 a classic overfitting signal \u2014 it&nbsp;<strong>automatically triggered pruning, removing redundant factors round by round<\/strong>.<\/li>\n\n\n\n<li>When it found multiple paths converging on the same set of core signals, it&nbsp;<strong>added multi-seed union validation<\/strong>&nbsp;to eliminate path dependence.<\/li>\n\n\n\n<li>When it judged that three-model ensembling was less robust than fixed-direction synthesis on small cross-sections, it&nbsp;<strong>autonomously switched to a more suitable strategy framework<\/strong>.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Breadth \u2014 massively parallel factor mining.<\/strong>&nbsp;Factor research entails a vast search space, and traditional workflows remain serial. Qwen3.8-Max parallelized the process: from just&nbsp;<strong>six short descriptions<\/strong>&nbsp;spanning the classic factor families of momentum, value, quality, investment, low-risk, and sentiment, it decomposed each into&nbsp;<strong>50 research directions<\/strong>, dispatched&nbsp;<strong>~330 sub-agents<\/strong>, completed&nbsp;<strong>~6,000 backtests<\/strong>, and continuously adapted the workflow mid-run. The selected factors achieved&nbsp;<strong>excess Sharpe ratios of 0.64\u20131.48<\/strong>, with IC uniformly positive, ranging from&nbsp;<strong>0.010 to 0.014<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">From coherent single-track R&amp;D to parallel exploration of a huge hypothesis space, Qwen3.8-Max leverages Dynamic Workflows to&nbsp;<strong>freeze orchestration logic into reproducible programs<\/strong>&nbsp;\u2014 compressing quant research that once took researchers&nbsp;<strong>weeks to months<\/strong>&nbsp;of serial work into a&nbsp;<strong>scalable, automated loop delivered within a single conversation<\/strong>, demonstrating the model\u2019s broad potential for&nbsp;<strong>long-horizon autonomous work<\/strong>.<\/p>\n\n\n\n<figure class=\"wp-block-video\"><video loop poster=\"https:\/\/img.alicdn.com\/imgextra\/i3\/O1CN01AwIkna24rPTFeFt88_!!6000000007444-0-tps-3200-1800.jpg\" src=\"https:\/\/cloud.video.taobao.com\/vod\/3lqSB8ahF_uFyftvfwYyHG2HTv0XmGsSWTbPjSLlbwg.mp4\" playsinline><\/video><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Video 2. Qwen3.8-Max promises to put a quant researcher&#8217;s expertise within everyone&#8217;s reach \u2014 showcasing the&nbsp;<em>depth<\/em>&nbsp;of its working ability.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Long-Horizon Task<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">When tackling highly complex, long-horizon, and multi-constraint tasks, Qwen3.8-Max demonstrates exceptional system-level autonomous planning and end-to-end closed-loop adaptive learning. Whether navigating stringent physical constraints in digital chip design or highly competitive, strategic business simulations, the model achieves deep algorithmic and strategic refactoring across thousands of rounds of interaction via an action-feedback-iteration loop.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Autonomous Chip Design and Closed-Loop Feedback-Driven Optimization<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen3.8-Max has independently achieved the autonomous execution of the entire silicon design flow, spanning logic restructuring, multi-constraint optimization, and physical layout generation. The target design is a&nbsp;<strong>GCD \/ RSA cryptographic hardware accelerator<\/strong>&nbsp;that integrates modular exponentiation and modular multiplication. Built on a GCD datapath and control path, this block represents a typically compact yet logic-dense digital circuit. Under a randomized&nbsp;<code>cocotb<\/code>&nbsp;verification framework, the model must maintain&nbsp;<strong>bit-exact functional correctness<\/strong>&nbsp;across 4-, 6-, 8-, and 16-bit configurations while minimizing the synthesized gate count (Yosys cell count)\u2014a direct addressing of the classic trade-off between area and correctness in front-end hardware design. Area performance is evaluated based on the 16-bit (WIDTH = 16) configuration.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen3.8-Max optimized this design within a sandboxed environment integrated with simulation (Iverilog), synthesis (Yosys), and physical design (OpenROAD) toolchains. Starting with minimal inputs\u2014a basic task description, a stub RTL workspace with empty module templates, and an evaluation script for verification and synthesis\u2014Qwen3.8-Max operated completely autonomously. Without any golden reference designs or human intervention, the model independently executed the entire process from high-level algorithmic architecture design to RTL code generation and multi-round iterative refinement.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Over a single continuous autonomous run, Qwen3.8-Max completed approximately&nbsp;<strong>500 turns and 71 evaluations across 13 key milestones<\/strong>, executing an end-to-end restructure of the design. The model autonomously managed RTL editing, simulation debugging, synthesis analysis, redundancy localization, and iterative datapath re-architecting\u2014advancing from initial bug-fixing to deep, algorithm-level rewrites.&nbsp;<strong>While its first functionally viable design measured 8,298 gates, Qwen3.8-Max drove this down to 678 gates, leading all evaluated models<\/strong>. This trajectory demonstrates that Qwen3.8-Max is capable of major structural breakthroughs even hundreds of turns into a run, rather than plateauing after early, low-hanging gains.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Key Design Milestones Along the Trajectory:<\/strong><em>(The evolution records preserve the complete circuit topology and the corresponding code diff details at each stage)<\/em><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Algorithmic Rewrite: Modulo divider to iterative shift-subtract (8,298 \u2192 2,010 gates, Turn 22)<\/strong>&nbsp;The single largest optimization step. Qwen3.8-Max replaced the expensive 16-bit hardware modulo divider in&nbsp;<code>modular_multiplier<\/code>&nbsp;with an iterative shift-subtract architecture, slashing 6,288 gates in one move\u2014accounting for over 80% of the total area reduction.<\/li>\n\n\n\n<li><strong>Redundancy Elimination &amp; Bitwidth Trimming (2,010 \u2192 1,304 gates, Turns 35\u201348)<\/strong>&nbsp;Recognizing the caller\u2019s pre-conditions, the model safely bypassed the entire&nbsp;<code>REDUCE<\/code>&nbsp;stage, merged two independent reduction modules into a single shared block, optimized the output path to combinational logic, and narrowed the bitwidth of the internal register&nbsp;<code>k_ff<\/code>.<\/li>\n\n\n\n<li><strong>Register &amp; Control FSM Pruning (1,304 \u2192 907 gates, Turns 60\u2013113)<\/strong>&nbsp;The model removed redundant&nbsp;<code>base<\/code>&nbsp;and&nbsp;<code>mod<\/code>&nbsp;registers as well as the&nbsp;<code>k_nz<\/code>&nbsp;flip-flop, introduced an early-exit mechanism for even numbers, utilized the subtractor\u2019s most significant bit (MSB) as the comparator, and merged the separate \u201ccompare-then-subtract\u201d logic in the GCD module into a single, reusable subtractor.<\/li>\n\n\n\n<li><strong>Module Fusion &amp; Logic Sharing (907 \u2192 765 gates, Turns 170\u2013252)<\/strong>&nbsp;Dissolving module boundaries, the model inlined the multiplier directly into the modular exponentiation&nbsp;<strong>finite state machine (FSM)<\/strong>, merged three sub-modules, and shared a single subtractor globally, thereby eliminating cross-module redundant interfaces and duplicated logic.<\/li>\n\n\n\n<li><strong>Gate-Level Refinement (765 \u2192 678 gates, Turns 443\u2013500)<\/strong>&nbsp;Utilizing local optimizations such as a shared NOR-gate tree, absolute-difference subtraction splitting (abs-sub splitting), and byte-to-bit selection logic, the model squeezed out the final gate-level redundancies.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">To verify whether front-end optimizations translate to physical implementation, Qwen3.8-Max ran the RTL design through a standard place-and-route (PR) flow using OpenROAD (Nangate45 PDK) to generate a physical silicon layout. In the physical layout representation, each chip demonstrates the actual routing results: standard cells are laid out on the physical plane of the die, with metal routing layers stacked above (each layer color-coded and connected by vertical vias).&nbsp;<strong>The starting design occupied a 106\u00d7106 \u00b5m\u00b2 die<\/strong>&nbsp;with a total wirelength of 33,369 \u00b5m and severe timing violations (a negative slack of -4.46 ns).&nbsp;<strong>The final layout shrank to a 46\u00d746 \u00b5m\u00b2 die<\/strong>, with wirelength dropping to 4,187 \u00b5m, and successfully achieved timing closure at 500 MHz (+0.66 ns Slack). This represents an 81% reduction in physical die area, proving that high-level front-end architectural optimizations translate directly into highly compact, routable, and performant silicon implementation.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Expand<\/p>\n\n\n\n<figure class=\"wp-block-embed\"><div class=\"wp-block-embed__wrapper\">\nhttps:\/\/docs.qwenlm.ai\/resources\/CCoLV_eda_agent_square.html\n<\/div><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">This case highlights two pivotal capabilities of Qwen3.8-Max as a foundational model for autonomous, long-horizon hardware agents:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>Long-horizon Sustained Optimization<\/strong>: The model maintains a highly coherent, systematic strategy over hundreds of complex interaction turns, driving deep into algorithmic-level datapath rewrites rather than stalling at superficial syntax adjustments.<\/li>\n\n\n\n<li><strong>Feedback-driven Closed-loop Improvement<\/strong>: In the absence of prior reference designs, the model relies entirely on an \u201cedit-simulate-synthesize-layout\u201d feedback loop to drive optimization. Each design iteration is strictly validated through automated&nbsp;<code>cocotb<\/code>&nbsp;functional tests, with physical feasibility fully guaranteed by OpenROAD backend validation.<\/li>\n<\/ol>\n\n\n\n<h3 class=\"wp-block-heading\">Continuous Learning in Long-term Operations<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">E-Commerce Bench is a&nbsp;<strong>365-day long-cycle e-commerce operation<\/strong>&nbsp;simulation benchmark, designed to evaluate large language models\u2019 business decision-making capabilities in sustained operational scenarios. Built on real, desensitized transaction data from Taobao and Tmall, this benchmark deeply replicates a complex ecosystem comprising&nbsp;<strong>12 store types, 60 product categories, nearly 600 suppliers, and 7,000 products<\/strong>. The model is given \u00a5100,000 in starting capital to simultaneously operate multiple online stores. Throughout the year, it must contend with seasonal demand swings, sudden environmental events, and cash flow pressures from a highly realistic e-commerce settlement system. The model must autonomously make full-chain decisions, including product selection, supply chain negotiation, inventory management, dynamic pricing, and returns handling, with the ultimate goal of maximizing total balance by year-end. This also tests the model\u2019s capital allocation strategy throughout the year. It must know when to invest proactively for growth. Just as importantly, it must convert inventory and operating gains into cash before the cycle ends. Otherwise, unconverted assets left on the books can hurt the final results.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">In price negotiations, the benchmark introduces&nbsp;<strong>a supplier matrix, driven by game theory principles<\/strong>, where each supplier possesses distinct personality traits and concession strategies. This requires the model to negotiate through multi-round natural language interactions. Qwen3.8-Max demonstrated continuous learning capability in negotiations. It conducted deep probing on the same products from the same suppliers, achieving progressive reductions in procurement prices and steady increases in profit round by round. This caused&nbsp;<strong>the negotiation efficiency (represented by the area in the radar chart) to continuously expand over time<\/strong>. Moreover, it effectively generalized this negotiation experience to similar products, while other models\u2019 negotiation efficiency generally hit a plateau in the mid-term.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Additionally, the model had to navigate hidden risks beneath the surface and complex market rhythms. Within the matrix of nearly 600 suppliers, the benchmark covertly embedded 152 fraudulent merchants, encompassing classic scam patterns such as \u201cmembership fee traps,\u201d \u201clow-price bait,\u201d and \u201cgoods not as described.\u201d This comprehensively tested the model\u2019s risk control capabilities. At the same time, the pressure of surging orders during annual major promotions intertwined with random supply chain crises, like typhoons and material shortages, pushing the model\u2019s stocking rhythm and crisis management abilities to the limit. Against this backdrop, Qwen3.8-Max exhibited exceptional forward-looking planning capability. It invested the most capital in the earliest stage of operations to establish its position, which accelerated its subsequent asset growth curve. It also achieved&nbsp;<strong>a net profit exceeding \u00a5100,000 during the year-end major promotion period<\/strong>\u2014nearly 2.4 times that of the second-place GLM 5.2.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen3.8-Max ultimately achieved the highest total balance of \u00a5416,252 (a 4.16x return), surpassing the second-place GLM 5.2 by 38%. This also represents a 152% improvement over its previous flagship generation, Qwen3.7-Max. These results demonstrate that Qwen3.8-Max possesses advantages in&nbsp;<strong>long-horizon coherent decision-making<\/strong>. Furthermore, it has the ability to&nbsp;<strong>adaptively learn from transactional feedback<\/strong>, continuously iterating and evolving across more than 2,000 rounds of interaction, rather than rigidly adhering to strategies learned early on.<\/p>\n\n\n\n<figure class=\"wp-block-video\"><video loop poster=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN01A4Psos1yegm94b5md_!!6000000006604-2-tps-6667-3751.png\" src=\"https:\/\/cloud.video.taobao.com\/vod\/c_n2NFWqHeC6isiHkMeVGt3Yc5gpkwXgJZjWPC82UKc.mp4\" playsinline><\/video><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">Multimodal Agents<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">From everything it sees to everything it does, Qwen3.8-Max is not merely capable of understanding images, documents, and videos. It delivers&nbsp;<strong>visual intelligence that runs through the entire task lifecycle<\/strong>.<\/p>\n\n\n\n<figure class=\"wp-block-video\"><video loop poster=\"https:\/\/img.alicdn.com\/imgextra\/i4\/O1CN01UtzwQd1fClvzURuYx_!!6000000003971-2-tps-6667-3751.png\" src=\"https:\/\/cloud.video.taobao.com\/vod\/u4AnWfxHc8OqrpId-i0vwk0kVgGM2cl3D5J2P4z5hO4.mp4\" playsinline><\/video><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">When working with financial reports and complex PDFs spanning&nbsp;<strong>more than 200 pages<\/strong>, Qwen3.8-Max can understand text, charts, and document layouts across pages, extract key insights from large volumes of information, and turn them into structured reports or production-ready web experiences. When processing videos longer than&nbsp;<strong>100 hours<\/strong>, it can do more than locate specific moments and answer detailed questions. It can organize people, events, timestamps, and scenes into a&nbsp;<strong>video memory graph<\/strong>, continuously building connections across long time spans to reconstruct event progressions, character relationships, and critical moments.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Whether the input is a hundreds-page document, a complete TV series, or a 100-hour livestream, information that would otherwise be difficult to consume can be transformed into a&nbsp;<strong>searchable, traceable, and interactive knowledge structure<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Beyond understanding, Qwen3.8-Max can carry out real visual production tasks. It can edit personal footage into a vlog, turn a question into an immersive educational animation, reconstruct a complete frontend project from a single interface screenshot, transform a floor plan into a Blender-based 3D interior visualization, and develop interactive games and applications from a natural-language request.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">More importantly,&nbsp;<strong>vision is not limited to the input stage<\/strong>. During execution, Qwen3.8-Max continuously observes and evaluates its own intermediate results. It can inspect page layouts, object orientations, spatial relationships, animation quality, and interaction outcomes. When it detects issues\u2014such as a television facing the wrong direction, a misaligned interface, or a visual result that does not match the intended design\u2014it can&nbsp;<strong>identify the deviation, revise its plan, and correct the output autonomously<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This means vision is no longer simply another modality that an agent uses to understand input. It becomes a&nbsp;<strong>native feedback loop across planning, execution, verification, and iteration<\/strong>. The model generates while observing, acts while reviewing, and repeatedly examines the result, identifies problems, and improves its work. This visual feedback loop moves an agent beyond merely completing a task toward&nbsp;<strong>completing it well<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen3.8-Max is helping multimodal agents evolve from&nbsp;<strong>understanding the world<\/strong>&nbsp;to&nbsp;<strong>continuously acting and creating within it through vision<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">In the digital world, finishing a complex task on its own often takes two things at once:&nbsp;<strong>writing code to implement the underlying logic, and operating the interface by hand to drive the task and observe the result.<\/strong>&nbsp;This Hybrid Agent capability \u2014 the pairing of&nbsp;<em>coding<\/em>&nbsp;and&nbsp;<em>GUI operation<\/em>&nbsp;\u2014 makes the two channels complementary:&nbsp;<strong>coding does the heavy lifting efficiently and at scale<\/strong>, while&nbsp;<strong>GUI operation reaches whatever a human can see and touch and, just as importantly, feeds back what actually happens in a live system<\/strong>&nbsp;\u2014 extending the visual feedback loop above from inspecting its own output to&nbsp;<strong>verifying against a real, running application<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">To measure this, we introduce&nbsp;<strong>RecreationBench<\/strong>, a long-horizon application-recreation benchmark spanning five platforms \u2014 desktop (Ubuntu, macOS, Windows), mobile (Android), and web. The model may observe a real, running application only as a&nbsp;<strong>black box<\/strong>&nbsp;\u2014 no source code, no internet access \u2014 making sense of it purely through interaction and feedback, then rebuilding the whole application from scratch. Here Qwen3.8-Max already demonstrates&nbsp;<strong>frontier-level Hybrid Agent capability<\/strong>, converging on the original step by step through repeated cycles of iterative coding and interactive feedback.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">To make these capabilities easier to integrate into existing agent systems, we are also introducing&nbsp;<strong>Qwen-MM-Plugins<\/strong>. It is a harness extension library designed for multimodal agents, providing agent frameworks with image and video processing, multimodal memory, dynamic-resolution support, visual tool use, and specialized capabilities for tasks such as video editing, Blender, and CAD. With Qwen-MM-Plugins,&nbsp;<strong>any existing agent harness can be extended into a more naturally multimodal-native system<\/strong>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">User Feedback<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">The most honest take on Qwen3.8-Max comes from people who actually put it to work. Top-tier agent platforms, leading open-source algorithm teams, professional firms in law, finance, and manufacturing, scrappy startups, solo developers, and academic researchers \u2014 all of them keep handing it their&nbsp;<strong>most complex, mission-critical, and long-horizon tasks<\/strong>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Enterprises use it to stand up large-scale agent systems. Knowledge workers dump their images, manuscripts, and video on it, and get everything processed. Developers hand it their heaviest engineering tasks outright. Research teams run the loop of literature, data, and simulation end to end. One model, reached for so often across such different work that it becomes indispensable. The verdict is the same:&nbsp;<strong>Qwen3.8-Max drives long, autonomous task chains and turns out ship-ready results in a single pass<\/strong>.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN013ZQbOo1uUNOW45Tt0_!!6000000006040-2-tps-3539-4096.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i3\/O1CN01acGhQo1dwSCtbsT62_!!6000000003800-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN01NMYOi52A4IhHdPIgL_!!6000000008149-2-tps-3240-3750.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i1\/O1CN01rX7sTc25Wd7IzpkWk_!!6000000007534-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i1\/O1CN01WsuXnb2AI2Zth3fQy_!!6000000008179-2-tps-3240-3750.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i3\/O1CN01HwDnic1zHcRPI05Pe_!!6000000006689-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN01H050So20tSmdPCjlc_!!6000000006907-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i3\/O1CN01JKaiTs1iVxN70XQLK_!!6000000004419-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i4\/O1CN01SmbYup1oSeCl6nced_!!6000000005224-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN01VO6xZr1pxiI6Zryg_!!6000000002571-2-tps-3240-3750.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN01Iur92d1hDLfEtN7ET_!!6000000004243-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN013ZQbOo1uUNOW45Tt0_!!6000000006040-2-tps-3539-4096.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i3\/O1CN01acGhQo1dwSCtbsT62_!!6000000003800-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN01NMYOi52A4IhHdPIgL_!!6000000008149-2-tps-3240-3750.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i1\/O1CN01rX7sTc25Wd7IzpkWk_!!6000000007534-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i1\/O1CN01WsuXnb2AI2Zth3fQy_!!6000000008179-2-tps-3240-3750.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i3\/O1CN01HwDnic1zHcRPI05Pe_!!6000000006689-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN01H050So20tSmdPCjlc_!!6000000006907-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i3\/O1CN01JKaiTs1iVxN70XQLK_!!6000000004419-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i4\/O1CN01SmbYup1oSeCl6nced_!!6000000005224-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN01VO6xZr1pxiI6Zryg_!!6000000002571-2-tps-3240-3750.png\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/img.alicdn.com\/imgextra\/i2\/O1CN01Iur92d1hDLfEtN7ET_!!6000000004243-2-tps-6480-7500.png\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">Full Benchmark Table<\/h2>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th class=\"has-text-align-left\" data-align=\"left\"><\/th><th class=\"has-text-align-center\" data-align=\"center\">Opus4.8<\/th><th class=\"has-text-align-center\" data-align=\"center\">Fable5<\/th><th class=\"has-text-align-center\" data-align=\"center\">GPT5.6 Sol (max)<\/th><th class=\"has-text-align-center\" data-align=\"center\">Qwen3.7-Max<\/th><th class=\"has-text-align-center\" data-align=\"center\">Qwen3.8-Max<\/th><\/tr><\/thead><tbody><tr><td class=\"has-text-align-left\" data-align=\"left\" colspan=\"6\">Coding Agent<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">Terminal Bench 2.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">88.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">74.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">SWE-bench Pro<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">64.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">60.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">67.7<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">DeepSWE 1.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">59.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">21.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">56.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">NL2Repo-Bench<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">47.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">55.9<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">FrontierSWE<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">88.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">40.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.5<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MLS-Bench-Lite<\/td><td class=\"has-text-align-center\" data-align=\"center\">42.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">49.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">46.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">31.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">41.0<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">PaperBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">88.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">90.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">64.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">93.0<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">AndroidBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">74.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">56.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">75.1<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">QwenSWEBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">63.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.7<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">QwenQoderBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">62.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">63.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">53.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">36.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">58.4<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">QwenReactBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">1694<\/td><td class=\"has-text-align-center\" data-align=\"center\">1770<\/td><td class=\"has-text-align-center\" data-align=\"center\">1564<\/td><td class=\"has-text-align-center\" data-align=\"center\">1538<\/td><td class=\"has-text-align-center\" data-align=\"center\">1724<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">QwenSVGBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">1648<\/td><td class=\"has-text-align-center\" data-align=\"center\">1690<\/td><td class=\"has-text-align-center\" data-align=\"center\">1758<\/td><td class=\"has-text-align-center\" data-align=\"center\">1499<\/td><td class=\"has-text-align-center\" data-align=\"center\">1713<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\" colspan=\"6\">General Agent<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">CoWorkBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">75.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">71.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">64.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">74.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">WorkSpaceBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">66.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">65.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">61.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">67.7<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">JobBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">48.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">57.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">45.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">31.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">53.4<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">SkillsBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">65.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">61.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.2<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">Agents&#8217; Last Exam (Pass \/ Score)<\/td><td class=\"has-text-align-center\" data-align=\"center\">27.0 \/ 45.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212; \/ &#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">30.6 \/ 53.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">11.8 \/ 31.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">27.0 \/ 52.4<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">Automation-Bench (Pass@1)<\/td><td class=\"has-text-align-center\" data-align=\"center\">27.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">29.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">29.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">14.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">27.3<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">Toolathlon Verified (Pass@1)<\/td><td class=\"has-text-align-center\" data-align=\"center\">76.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">74.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">49.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.5<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">WideSearch<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">75.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.9<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">HLE w\/ tools<\/td><td class=\"has-text-align-center\" data-align=\"center\">57.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">64.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">58.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">53.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">56.2<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\" colspan=\"6\">General Capabilities<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">GPQA Diamond<\/td><td class=\"has-text-align-center\" data-align=\"center\">92.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">92.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">94.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">92.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">92.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">HLE<\/td><td class=\"has-text-align-center\" data-align=\"center\">45.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">53.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">47.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">41.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">43.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">IFBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">62.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">63.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">79.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">$OneMillion-Bench (expert score)<\/td><td class=\"has-text-align-center\" data-align=\"center\">41.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">55.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">53.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">44.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">52.5<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">HealthBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">52.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">55.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">54.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">60.2<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">PLawBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">58.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.2<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">PRBench-Legal<\/td><td class=\"has-text-align-center\" data-align=\"center\">52.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">57.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">57.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">48.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">57.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">PRBench-Finance<\/td><td class=\"has-text-align-center\" data-align=\"center\">51.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">55.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">55.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">46.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">58.3<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MRCR v2 256K (8-needle)<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">93.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">92.9<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">LongBench v2<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">67.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">65.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">66.3<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">1. Fable5 results may involve fallbacks.<br>2. Terminal Bench 2.1: Evaluated with Claude Code (avg@10), using a 5-hour timeout and max_tokens=131,072. For all other models, we report the best published score across harnesses: Claude Opus 4.8 and Claude Fable 5 with Terminus 2 from Artificial Analysis (https:\/\/artificialanalysis.ai\/evaluations\/terminalbench-v2-1); GPT-5.6 Sol with Codex (https:\/\/openai.com\/index\/previewing-gpt-5-6-sol\/).<br>3. SWE-bench Pro: Evaluated with the Claude Code harness, temp=1.0, top_p=0.95, and a 256K context window. Problematic tasks corrected and all baselines evaluated on the refined benchmark.<br>4. DeepSWE 1.1: Evaluated with the Claude Code and mini-SWE-agent harnesses, temp=1.0, top_p=0.95, and a 256K context window. We report the highest score among both harnesses; notably, Qwen3.8-Max performs best on Claude Code.<br>5. NL2Repo-Bench: Evaluated with the Claude Code harness. To prevent reward hacking, we disable Bash commands that attempt to access the specific repository, such as pip download, pip install, and git clone.<br>6. FrontierSWE: Evaluated with the Claude Code harness. All other available MEAN@5 results are taken from the official FrontierSWE leaderboard (https:\/\/www.frontierswe.com) as of August 3, 2026. Dominance scores are recomputed from the raw scores using the official evaluation script. &#8220;&#8211;&#8221; indicates that no official MEAN@5 result was available as of that date.<br>7. MLS-Bench-Lite: Evaluated with Claude Code using a 5-hour timeout and max_tokens=131,072. All other model scores are taken from the official leaderboard.<br>8. PaperBench: Evaluated in the BasicAgent setting under Code-Dev mode, judged by Claude Opus 4.6, and averaged over 3 runs (max 12 hours per run).<br>9. AndroidBench: Evaluated on the 95-task public subset, reporting avg@3 scores.<br>10. QwenSWEBench: Inhouse coding benchmark to evaluate models&#8217; software engineering capabilities. Evaluated with the Claude Code harness. Reporting avg@3 with an 8-hour timeout, max_tokens=32,768, temperature=1.0, and a 256K-token context window.<br>11. QwenQoderBench: Inhouse coding benchmark to evaluate user experience on Qoder. Evaluated with the Claude Code harness. Reporting avg@5 with a 6-hour timeout, max_tokens=32,768, temperature=1.0, and a 256K-token context window.<br>12. QwenReactBench: Inhouse React project building benchmark using Claude Code as the harness, bilingual (EN\/CN), 7 categories; auto-render + multimodal judge; BT\/Elo rating.<br>13. QwenSVGBench: Inhouse SVG code generation benchmark; bilingual (EN\/CN), auto-render + multimodal judge; BT\/Elo rating.<br>14. CoWorkBench: Inhouse cowork benchmark for evaluating long-horizon tasks across computer science, finance, law, medical, and other productivity domains.<br>15. SkillsBench: Evaluated on the public SkillsBench v1.1 benchmark across 87 tasks, reporting the average score over three runs per task. Opus 4.8 and Fable 5 are evaluated on Claude Code; GPT-5.6 Sol is evaluated on Codex; the Qwen-series are evaluated on OpenCode. All results are from our own testing.<br>16. Automation-Bench: Evaluated on the 600-task public subset.<br>17. WideSearch: Evaluated with the Claude Code harness for external models and the Qwen-Agent harness for ours, reporting the average item-F1 over four runs.<br>18. $OneMillion-Bench: Evaluated using gemini-3.1-pro-preview.<br>19. PLawBench: Evaluated using gemini-3.1-pro-preview.<br>20. Empty cells (&#8211;): Scores are not yet available or are not applicable.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th class=\"has-text-align-left\" data-align=\"left\"><\/th><th class=\"has-text-align-center\" data-align=\"center\">Opus4.8<\/th><th class=\"has-text-align-center\" data-align=\"center\">Fable5<\/th><th class=\"has-text-align-center\" data-align=\"center\">Gemini3.1-Pro<\/th><th class=\"has-text-align-center\" data-align=\"center\">GPT5.6-Sol<\/th><th class=\"has-text-align-center\" data-align=\"center\">Qwen3.7-Plus<\/th><th class=\"has-text-align-center\" data-align=\"center\">Qwen3.8-Max<\/th><\/tr><\/thead><tbody><tr><td class=\"has-text-align-left\" data-align=\"left\" colspan=\"7\">Multimodal Reasoning<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MMMU-Pro<\/td><td class=\"has-text-align-center\" data-align=\"center\">75.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">79.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.3<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MathVision<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.1 \/ 97.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">92.7 \/ 98.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.4 \/ 95.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">90.8 \/ 97.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">90.3 \/ &#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">95.2 \/ 97.7<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">BabyVision<\/td><td class=\"has-text-align-center\" data-align=\"center\">28.4 \/ 81.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">42.5 \/ 90.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">55.9 \/ 68.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">65.5 \/ 88.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">64.7 \/ 70.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.0 \/ 91.3<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">HLE-VL (w\/ Tools)<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">43.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">51.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">25.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">52.2<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">ZeroBench (Pass@5)<\/td><td class=\"has-text-align-center\" data-align=\"center\">17.0 \/ 34.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">20.0 \/ 46.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">17.0 \/ 23.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">22.0 \/ 35.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">19.0 \/ 19.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">24.0 \/ 49.0<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">ZeroBench-Sub<\/td><td class=\"has-text-align-center\" data-align=\"center\">31.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">37.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">36.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">46.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">41.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">48.5<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">LogicVista<\/td><td class=\"has-text-align-center\" data-align=\"center\">76.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">89.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">91.9<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">HiPhO<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">78.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">90.0<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">PhyX<\/td><td class=\"has-text-align-center\" data-align=\"center\">54.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">71.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">79.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">79.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.5<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">SLAKE<\/td><td class=\"has-text-align-center\" data-align=\"center\">75.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">90.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MedXpertQA-MM<\/td><td class=\"has-text-align-center\" data-align=\"center\">71.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">71.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.4<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">PMC-VQA<\/td><td class=\"has-text-align-center\" data-align=\"center\">59.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">63.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">62.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">62.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">63.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">66.2<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\" colspan=\"7\">Visual Agent &amp; Coding<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">OSWorld-Verified<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">76.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.1<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">OSWorld 2.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">20.6 \/ 54.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212; \/ 66.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">7.8 \/ 30.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212; \/ 62.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">2.8 \/ 21.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">19.4 \/ 46.7<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">ScreenSpot Pro<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">79.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.5<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">WebArena-Verified<\/td><td class=\"has-text-align-center\" data-align=\"center\">67.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">71.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">64.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">55.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">66.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">AndroidWorld<\/td><td class=\"has-text-align-center\" data-align=\"center\">75.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">88.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.3<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MobileWorld<\/td><td class=\"has-text-align-center\" data-align=\"center\">67.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">58.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">76.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">51.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">ClawEval-MM<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.3 \/ 73.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.2 \/ 77.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">50.5 \/ 55.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.2 \/ 78.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">57.4 \/ 60.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.2 \/ 74.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">Vision2Web<\/td><td class=\"has-text-align-center\" data-align=\"center\">62.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">62.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">42.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.0<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">QwenBlenderBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">62.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">23.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">41.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.9<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">Parametric CAD Bench<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">91.5<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">RecreationBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">48.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">56.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">16.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">47.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">30.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">51.7<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">PresentBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">79.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">55.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">65.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">79.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\" colspan=\"7\">Document &amp; Office Intelligence<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">CharXiv (RQ)<\/td><td class=\"has-text-align-center\" data-align=\"center\">78.5 \/ 89.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.9 \/ 93.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.4 \/ 89.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.1 \/ 89.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.8 \/ 85.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">88.4 \/ 93.5<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">OmniDocBench 1.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">89.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">90.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">91.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">92.1<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">OCR-Bench-V2 (EN\/ZH)<\/td><td class=\"has-text-align-center\" data-align=\"center\">53.9 \/ 55.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">65.3 \/ 58.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">64.6 \/ 58.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.0 \/ 57.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.7 \/ 67.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">74.2 \/ 68.3<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">CC-OCR-Bench-V2<\/td><td class=\"has-text-align-center\" data-align=\"center\">60.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">79.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MTVQA-Test<\/td><td class=\"has-text-align-center\" data-align=\"center\">48.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">41.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">54.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">52.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">51.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">56.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MADQA<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">91.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">QwenVisualOffice<\/td><td class=\"has-text-align-center\" data-align=\"center\">34.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">32.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">39.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">29.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">32.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">44.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\" colspan=\"7\">Real-World &amp; Spatial Understanding<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">RealWorldQA<\/td><td class=\"has-text-align-center\" data-align=\"center\">76.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">88.0<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">ERQA<\/td><td class=\"has-text-align-center\" data-align=\"center\">57.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">LingoQA<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">66.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">SURDS<\/td><td class=\"has-text-align-center\" data-align=\"center\">62.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">79.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">64.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">63.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\" colspan=\"7\">Visual Perception &amp; Grounding<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">SimpleVQA<\/td><td class=\"has-text-align-center\" data-align=\"center\">67.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">66.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">75.0<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">WorldVQA<\/td><td class=\"has-text-align-center\" data-align=\"center\">33.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">53.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">54.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">45.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">43.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">53.2<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MMStar<\/td><td class=\"has-text-align-center\" data-align=\"center\">76.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.9<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">PerceptionBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">47.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">57.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">56.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">59.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">51.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">63.5<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">CountQA<\/td><td class=\"has-text-align-center\" data-align=\"center\">41.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">63.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.4<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">RefAdv-S<\/td><td class=\"has-text-align-center\" data-align=\"center\">61.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">71.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.2<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">Dense200<\/td><td class=\"has-text-align-center\" data-align=\"center\">20.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">31.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">69.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">55.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">60.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.0<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">COCO<\/td><td class=\"has-text-align-center\" data-align=\"center\">50.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">56.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">61.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">74.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">78.7<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">VisFactor<\/td><td class=\"has-text-align-center\" data-align=\"center\">30.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">54.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">39.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">62.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">42.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">60.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">VLMsAreBiased<\/td><td class=\"has-text-align-center\" data-align=\"center\">43.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">61.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">74.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">59.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">36.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">88.3<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\" colspan=\"7\">Video Intelligence &amp; Agents<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">VideoMME (w\/ Sub.)<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">86.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">89.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">88.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">90.4<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">VideoMME v2 (w\/ Sub.)<\/td><td class=\"has-text-align-center\" data-align=\"center\">49.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">52.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">66.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">71.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">59.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.3<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">VideoMMMU<\/td><td class=\"has-text-align-center\" data-align=\"center\">75.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">88.7<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MMVU<\/td><td class=\"has-text-align-center\" data-align=\"center\">67.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">72.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.9<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">76.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.4<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">MLVU (M-Avg)<\/td><td class=\"has-text-align-center\" data-align=\"center\">53.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.7<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">87.4<\/td><td class=\"has-text-align-center\" data-align=\"center\">90.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">TVBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">61.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">83.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">78.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.9<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">LVBench<\/td><td class=\"has-text-align-center\" data-align=\"center\">67.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">75.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">78.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">76.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">81.8<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">LVBench (w\/ Mem.)<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">90.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">84.2<\/td><td class=\"has-text-align-center\" data-align=\"center\">74.5<\/td><td class=\"has-text-align-center\" data-align=\"center\">85.6<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">EgoLife (w\/ Mem.)<\/td><td class=\"has-text-align-center\" data-align=\"center\">78.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">82.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">70.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">68.8<\/td><td class=\"has-text-align-center\" data-align=\"center\">80.3<\/td><\/tr><tr><td class=\"has-text-align-left\" data-align=\"left\">VideoDR (w\/ Search)<\/td><td class=\"has-text-align-center\" data-align=\"center\">65.6<\/td><td class=\"has-text-align-center\" data-align=\"center\">77.1<\/td><td class=\"has-text-align-center\" data-align=\"center\">&#8212;<\/td><td class=\"has-text-align-center\" data-align=\"center\">71.3<\/td><td class=\"has-text-align-center\" data-align=\"center\">41.0<\/td><td class=\"has-text-align-center\" data-align=\"center\">73.2<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">1. MathVision, BabyVision, CharXiv (RQ), and ZeroBench: Scores are reported as \u201cwithout CI \/ with CI.\u201d A small number of incorrect ground-truth annotations in MathVision and CharXiv (RQ) were corrected following manual verification.<br>2. MathVision: Our model is evaluated using a fixed prompt, e.g., \u201cPlease reason step by step, and put your final answer within&nbsp;<code>\\boxed{}<\/code>.\u201d For other models, we report the higher score obtained from runs with and without the&nbsp;<code>\\boxed{}<\/code>&nbsp;formatting requirement.<br>3. MMMU-Pro: Results for Gemini3.1-Pro and GPT5.6-Sol are taken from official model reports or system cards. All other models are evaluated in-house.<br>4. ClawEval-MM: Scores are reported as \u201cPass@3 \/ average score.\u201d Pass@3 measures the percentage passed in at least one of the three trials, and average score is the mean score across the three trials.<br>5. Vision2Web: Scores are averaged across the frontend, webpage, and website categories, using the Claude Code harness and gpt-5.4-2026-03-05 as the judge.<br>6. HLE-VL (w\/ Tools): Scores are evaluated with tool use, including both Code Interpreter (CI) and Search. Scores for the tool-enabled versions of Gemini3.1-Pro and GPT5.6-Sol are measured end-to-end through their official native tool-calling APIs.<br>7. OSWorld 2.0: Scores are reported as \u201cbinary \/ partial.\u201d The binary score is the percentage of tasks receiving the full task reward, while the partial score aggregates the partial rewards obtained across all tasks.<br>8. ScreenSpot Pro: Scores for Opus4.8 and Fable5 are taken from official system cards. The Fable5 results refer to the corresponding Mythos Preview scores. All other models are evaluated in-house.<br>9. WebArena-Verified: Scores are reported using the official WebArena grader within the OSWorld scaffold.<br>10. RecreationBench: An internal long-horizon application-recreation benchmark for evaluating hybrid-agent capabilities across five platforms: Ubuntu, macOS, Windows, Android, and the web.<br>11. PerceptionBench: Scores for comparison models are taken from the benchmark\u2019s official release report, while our model is evaluated in-house.<br>12. VideoMME (w\/ Sub.) and VideoMME v2 (w\/ Sub.): Scores are evaluated with subtitles enabled.<br>13. QwenBlenderBench and QwenVisualOffice: Both are internal benchmarks.<br>14. LVBench and EgoLife (w\/ Mem.): Scores are evaluated using a memory system built with Qwen-MM-Plugins, enabling fine-grained, long-horizon video memory.<br>15. VideoDR (w\/ Search): Scores are evaluated with access to a search tool.<br>16. Empty cells (&#8211;): Scores are not yet available or are not applicable.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Build with Qwen3.8<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen3.8-Max is now available through&nbsp;<a href=\"https:\/\/www.qwencloud.com\/\">QwenCloud<\/a>. You can integrate it with popular agent frameworks and coding assistants. The model weights will be open-sourced on Hugging Face and ModelScope next week \u2014 stay tuned.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">API Usage<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen3.8-Max comes with the official support for&nbsp;<code>reasoning_effort<\/code>, which can be used to adjust reasoning depth and control cost:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>xhigh<\/code>&nbsp;(default): for complex tasks demanding thorough analysis<\/li>\n\n\n\n<li><code>medium<\/code>: balancing accuracy and speed<\/li>\n\n\n\n<li><code>low<\/code>: efficient reasoning optimizing for speed and cost<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">In addition,&nbsp;<code>preserve_thinking<\/code>&nbsp;is enabled by default for all workloads for best out-of-the-box experience.<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">QwenCloud<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">QwenCloud supports industry-standard protocols, including chat completions and responses APIs compatible with OpenAI\u2019s specification, as well as an API interface compatible with Anthropic.<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">python<code>\"\"\"\nEnvironment variables:\n  DASHSCOPE_API_KEY: Your API Key from https:\/\/home.qwencloud.com\/\n  DASHSCOPE_BASE_URL: (optional) Base URL for compatible-mode API.\n    - Beijing: https:\/\/dashscope.aliyuncs.com\/compatible-mode\/v1\n    - Singapore: https:\/\/dashscope-intl.aliyuncs.com\/compatible-mode\/v1\n    - US (Virginia): https:\/\/dashscope-us.aliyuncs.com\/compatible-mode\/v1\n\"\"\"from openai import OpenAIimport os<br>api_key = os.environ.get(\"DASHSCOPE_API_KEY\")if not api_key:    raise ValueError(        \"DASHSCOPE_API_KEY is required. \"        \"Set it via: export DASHSCOPE_API_KEY='your-api-key'\"    )<br>client = OpenAI(    api_key=api_key,    base_url=os.environ.get(        \"DASHSCOPE_BASE_URL\",        \"https:\/\/dashscope-intl.aliyuncs.com\/compatible-mode\/v1\",    ),)<br>messages = [{\"role\": \"user\", \"content\": \"Write a Python function to merge two sorted linked lists.\"}]<br>completion = client.chat.completions.create(    model=\"qwen3.8-max\",    messages=messages,    extra_body={        \"enable_thinking\": True,        # \"preserve_thinking\": True,    },    reasoning_effort=\"xhigh\",  # supported levels are xhigh, medium, and low    stream=True,)<br>reasoning_content = \"\"answer_content = \"\"is_answering = Falseprint(\"\\n\" + \"=\" * 20 + \"Reasoning\" + \"=\" * 20 + \"\\n\")<br>for chunk in completion:    if not chunk.choices:        print(\"\\nUsage:\")        print(chunk.usage)        continue<br>    delta = chunk.choices[0].delta<br>    if hasattr(delta, \"reasoning_content\") and delta.reasoning_content is not None:        if not is_answering:            print(delta.reasoning_content, end=\"\", flush=True)        reasoning_content += delta.reasoning_content<br>    if hasattr(delta, \"content\") and delta.content:        if not is_answering:            print(\"\\n\" + \"=\" * 20 + \"Answer\" + \"=\" * 20 + \"\\n\")            is_answering = True        print(delta.content, end=\"\", flush=True)        answer_content += delta.content<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">For more information, please visit the&nbsp;<a href=\"https:\/\/docs.qwencloud.com\/developer-guides\/getting-started\/first-api-call\">API doc<\/a>.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Coding Assistants<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen3.8-Max integrates seamlessly with popular agent frameworks and coding assistants:<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">Claude Code<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen APIs support the Anthropic API protocol, enabling direct use with&nbsp;<strong>Claude Code<\/strong>:<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">bash<code>npm install -g @anthropic-ai\/claude-code\n<br>export ANTHROPIC_MODEL=\"qwen3.8-max\"export ANTHROPIC_SMALL_FAST_MODEL=\"qwen3.8-max\"export ANTHROPIC_BASE_URL=https:\/\/dashscope-intl.aliyuncs.com\/apps\/anthropic\nexport ANTHROPIC_AUTH_TOKEN=&lt;your_api_key&gt;\n<br>claude\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">Codex<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen APIs support the OpenAI Responses protocol, enabling use with&nbsp;<strong>Codex<\/strong>:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">In&nbsp;<code>~\/.codex\/model-catalog.local.json<\/code><\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">json<code>{  \"models\": [    {      \"slug\": \"qwen3.8-max\",      \"display_name\": \"qwen3.8-max\",      \"description\": \"Model Studio: Qwen3.8-Max\",      \"default_reasoning_level\": \"xhigh\",      \"supported_reasoning_levels\": [        {          \"effort\": \"low\",          \"description\": \"Fast responses with lighter reasoning\"        },        {          \"effort\": \"medium\",          \"description\": \"Greater reasoning depth for complex problems\"        },        {          \"effort\": \"xhigh\",          \"description\": \"Extra high reasoning depth for complex problems\"        }      ],      \"context_window\": 1000000,      \"effective_context_window_percent\": 95,      \"supports_parallel_tool_calls\": true,      \"supports_image_detail_original\": true,      \"input_modalities\": [\"text\", \"image\"],      \"shell_type\": \"default\",      \"visibility\": \"list\",      \"supported_in_api\": true,      \"priority\": 1,      \"base_instructions\": \"\",      \"support_verbosity\": false,      \"supports_reasoning_summaries\": false,      \"experimental_supported_tools\": [],      \"truncation_policy\": {        \"mode\": \"bytes\",        \"limit\": 10000      }    }  ]}<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">In&nbsp;<code>~\/.codex\/config.toml<\/code><\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">toml<code>model_catalog_json = \"~\/.codex\/model-catalog.local.json\"<br>model_provider = \"ModelStudio\"model = \"qwen3.8-max\"<br>[model_providers.ModelStudio]name = \"Model Studio\"base_url = \"https:\/\/dashscope-intl.aliyuncs.com\/compatible-mode\/v1\"env_key = \"OPENAI_API_KEY\"wire_api = \"responses\"<\/code><\/pre>\n\n\n\n<pre class=\"wp-block-preformatted\">bash<code>npm install -g @openai\/codex\n<br>export OPENAI_API_KEY=&lt;your_api_key&gt;\n<br>codex\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">Qoder CLI<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/qoder.com\/\">Qoder<\/a>&nbsp;co-evolves with Qwen for agentic coding:<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">bash<code>curl -fsSL https:\/\/qoder.com\/install | bash\n<br>qoder\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">Qwen Code<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/qwen.ai\/qwencode\">Qwen Code<\/a>&nbsp;is deeply optimized for the Qwen series:<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">bash<code>npm install -g @qwen-code\/qwen-code@latest\n<br>qwen\n<\/code><\/pre>\n\n\n\n<h4 class=\"wp-block-heading\">OpenClaw<\/h4>\n\n\n\n<p class=\"wp-block-paragraph\">Connect to&nbsp;<a href=\"https:\/\/openclaw.ai\/\">OpenClaw<\/a>&nbsp;via&nbsp;<a href=\"https:\/\/docs.qwencloud.com\/developer-guides\/clients-and-developer-tools\/openclaw\">QwenCloud<\/a>:<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">bash<code>curl -fsSL https:\/\/molt.bot\/install.sh | bash\n<br>export DASHSCOPE_API_KEY=&lt;your_api_key&gt;\n<br>openclaw dashboard\n<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">Configure&nbsp;<code>~\/.openclaw\/openclaw.json<\/code>:<\/p>\n\n\n\n<pre class=\"wp-block-preformatted\">json<code>{  \"models\": {    \"mode\": \"merge\",    \"providers\": {      \"modelstudio\": {        \"baseUrl\": \"https:\/\/dashscope-intl.aliyuncs.com\/compatible-mode\/v1\",        \"apiKey\": \"DASHSCOPE_API_KEY\",        \"api\": \"openai-completions\",        \"models\": [          {            \"id\": \"qwen3.8-max\",            \"name\": \"qwen3.8-max\",            \"reasoning\": true,            \"input\": [\"text\", \"image\"],            \"contextWindow\": 1000000,            \"maxTokens\": 65536          }        ]      }    }  },  \"agents\": {    \"defaults\": {      \"model\": {        \"primary\": \"modelstudio\/qwen3.8-max\"      }    }  }}<\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\">Summary<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Qwen3.8-Max is our most capable model to date, and the first open-weight model at Max scale. Scaling to 2.4 trillion parameters, it delivers comprehensive gains across coding, real-world work, long-horizon tasks, and multimodal agents \u2014 able to take complex, open-ended goals from start to finish with minimal human involvement and produce dependable deliverables. The open weights will be released next week. We welcome community feedback and look forward to seeing what you build.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Today, we are officially releasing&nbsp;Qwen 3.8-Max, the most capable model in the Qwen family to date. This also marks the first time we will open-source [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":286,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[],"tags":[],"class_list":["post-226","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry"],"_links":{"self":[{"href":"https:\/\/tongyi.ai\/index.php?rest_route=\/wp\/v2\/posts\/226","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/tongyi.ai\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/tongyi.ai\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/tongyi.ai\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/tongyi.ai\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=226"}],"version-history":[{"count":2,"href":"https:\/\/tongyi.ai\/index.php?rest_route=\/wp\/v2\/posts\/226\/revisions"}],"predecessor-version":[{"id":284,"href":"https:\/\/tongyi.ai\/index.php?rest_route=\/wp\/v2\/posts\/226\/revisions\/284"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/tongyi.ai\/index.php?rest_route=\/wp\/v2\/media\/286"}],"wp:attachment":[{"href":"https:\/\/tongyi.ai\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=226"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/tongyi.ai\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=226"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/tongyi.ai\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=226"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}