Qwen3.8-2.4T-A95B 开源，首次开放 Max 级模型
Qwen3.8-2.4T-A95B 模型发布
Qwen 首次开源 Max 级模型，2.4T 总参、95B 激活，Agent 基准多项超 Claude 与 GPT，做 Agent 和长程任务的团队值得立刻上手实测。
Qwen3.8-2.4T-A95B
Qwen3.8-2.4T-A95B
This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format.
此仓库包含以 Hugging Face Transformers 格式存储的后训练模型权重和配置文件。
These artifacts are compatible with vLLM, SGLang, TokenSpeed, etc.
这些工件与 vLLM、SGLang、TokenSpeed 等兼容。
For users seeking managed, scalable inference without infrastructure maintenance, the official Qwen API service is provided by Qwen Cloud.
对于寻求托管、可扩展推理且无需维护基础设施的用户，官方 Qwen API 服务由 Qwen Cloud 提供。
In particular, Qwen3.8-Max is the official version based on Qwen3.8-2.4T-A95B with more features, such as vision input & non-thinking support, 1M context length by default, official built-in tools, etc. For more information, please refer to the Qwen3.8-Max Overview.
特别是，Qwen3.8-Max 是基于 Qwen3.8-2.4T-A95B 的官方版本，具有更多功能，例如视觉输入和非思考支持、默认 1M 上下文长度、官方内置工具等。更多信息请参阅 Qwen3.8-Max 概述。
Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date.
继 Qwen3.5 和 Qwen3.6 系列被社区广泛采用之后，我们很高兴推出 Qwen3.8，这是迄今为止 Qwen 开源模型家族中最强大的版本。
For the first time, Qwen3.8 brings a Qwen-Max-class model to open release. Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Beyond answering harder questions, Qwen3.8 is designed to carry complex, multi-step tasks through to completion with greater reliability.
Qwen3.8 首次将 Qwen-Max 级别的模型开源发布。基于 Qwen3.5 的架构基础，Qwen3.8 在编码、专业工作、研究和长周期智能体任务方面带来了显著提升。除了回答更难的问题，Qwen3.8 还旨在以更高的可靠性完成复杂的多步骤任务。
Qwen3.8 Highlights
Qwen3.8 亮点
Qwen3.8 features the following enhancements:
Qwen3.8 具有以下增强功能：
- Core Capabilities: Comprehensive improvements across coding, professional work, research, and long-horizon agentic tasks.
- Agent Execution: Stronger autonomous planning and better handling of environment feedback, leading to more reliable end-to-end task completion.
- Downstream Compatibility: Broader support for popular harnesses and development tools, making it easier to integrate into your existing stack.
- Flexible Thinking Control: Reasoning depth can be tuned with reasoning_effort, and reasoning context from historical messages is retained via preserve_thinking.
- 核心能力：在编码、专业工作、研究和长周期智能体任务方面全面改进。
- 智能体执行：更强的自主规划能力和更好的环境反馈处理，从而实现更可靠的端到端任务完成。
- 下游兼容性：更广泛地支持流行的测试框架和开发工具，使其更容易集成到您现有的技术栈中。
- 灵活的思考控制：可以通过 reasoning_effort 调整推理深度，并通过 preserve_thinking 保留历史消息中的推理上下文。
For more details, please refer to our blog post Qwen3.8-Max.
更多详情，请参阅我们的博客文章 Qwen3.8-Max。
Model Overview
模型概述
- Type: Causal Language Model
- Training Stage: Pre-training & Post-training
- Language Model
- Number of Parameters: 2.4T in total and 95B activated
- Hidden Dimension: 8192
- Token Embedding: 248,320 (Padded)
- Number of Layers: 92
- Hidden Layout: 23 × (3 × (Gated DeltaNet → MoE) → 1 × (Gated Attention → MoE))
- Gated DeltaNet:
- Number of Linear Attention Heads: 128 for V and 16 for QK
- Head Dimension: 128
- Gated Attention:
- Number of Attention Heads: 64 for Q and 4 for KV
- Head Dimension: 256
- Rotary Position Embedding Dimension: 64
- Mixture of Experts:
- Number of Experts: 512
- Number of Activated Experts: 10 Routed + 1 Shared
- Expert Intermediate Dimension: 2048
- LM Output: 248,320 (Padded)
- MTP (Multi-Token Prediction): trained with multiple steps
- Context Length: 262,144 natively and extensible up to 1,010,000 tokens.
- 类型：因果语言模型
- 训练阶段：预训练和后训练
- 语言模型
- 参数数量：总计 2.4T，激活 95B
- 隐藏维度：8192
- 词嵌入：248,320（填充后）
- 层数：92
- 隐藏布局：23 × (3 × (Gated DeltaNet → MoE) → 1 × (Gated Attention → MoE))
- Gated DeltaNet：
- 线性注意力头数量：V 为 128，QK 为 16
- 头维度：128
- 门控注意力：
- 注意力头数量：Q 为 64，KV 为 4
- 头维度：256
- 旋转位置嵌入维度：64
- 混合专家：
- 专家数量：512
- 激活专家数量：10 个路由 + 1 个共享
- 专家中间维度：2048
- LM 输出：248,320（填充后）
- MTP（多令牌预测）：通过多步训练
- 上下文长度：原生 262,144，可扩展至 1,010,000 个令牌。
Benchmark Results
基准测试结果
| Opus 4.8 | Fable 5 | GPT 5.6 Sol (max) | Qwen3.7-Max | Qwen3.8-Max | |
|---|---|---|---|---|---|
| Coding Agent | |||||
| Terminal Bench 2.1 | 84.6 | 84.6 | 88.8 | 74.5 | 86.6 |
| SWE-bench Pro | 69.2 | 80.0 | 64.6 | 60.6 | 67.7 |
| DeepSWE 1.1 | 59.0 | 70.0 | 73.0 | 21.6 | 56.6 |
| NL2Repo-Bench | 69.4 | -- | -- | 47.2 | 55.9 |
| FrontierSWE | 70.0 | 88.8 | -- | 40.7 | 73.5 |
| MLS-Bench-Lite | 42.8 | 49.9 | 46.2 | 31.7 | 41.0 |
| PaperBench | 80.3 | 88.8 | 90.5 | 64.8 | 93.0 |
| AndroidBench | 69.8 | 84.5 | 74.0 | 56.5 | 75.1 |
| QwenSWEBench | 84.0 | 86.3 | 73.5 | 63.4 | 80.7 |
| QwenQoderBench | 62.7 | 63.1 | 53.8 | 36.8 | 58.4 |
| QwenReactBench | 1694 | 1770 | 1564 | 1538 | 1724 |
| QwenSVGBench | 1648 | 1690 | 1758 | 1499 | 1713 |
| General Agent | |||||
| CoWorkBench | 72.3 | 75.9 | 71.5 | 64.6 | 74.8 |
| WorkSpaceBench | 66.8 | 68.7 | 65.6 | 61.4 | 67.7 |
| JobBench | 48.4 | 57.4 | 45.4 | 31.3 | 53.4 |
| SkillsBench | 65.1 | 70.9 | 73.5 | 61.2 | 70.2 |
| Agents' Last Exam (Pass / Score) | 27.0 / 45.1 | -- / -- | 30.6 / 53.6 | 11.8 / 31.1 | 27.0 / 52.4 |
| Automation-Bench (Pass@1) | 27.2 | 29.1 | 29.7 | 14.2 | 27.3 |
| Toolathlon Verified (Pass@1) | 76.2 | 77.9 | 74.9 | 49.7 | 72.5 |
| WideSearch | 72.9 | 81.2 | -- | 75.2 | 81.9 |
| HLE w/ tools | 57.9 | 64.5 | 58.0 | 53.5 | 56.2 |
| General Capabilities | |||||
| GPQA Diamond | 92.0 | 92.6 | 94.1 | 92.4 | 92.6 |
| HLE | 45.7 | 53.3 | 47.2 | 41.4 | 43.6 |
| IFBench | 62.2 | 63.5 | 72.7 | 79.1 | 82.8 |
| $OneMillion-Bench (expert score) | 41.8 | 55.9 | 53.8 | 44.4 | 52.5 |
| HealthBench | 52.4 | -- | 55.3 | 54.5 | 60.2 |
| PLawBench | 69.6 | 70.2 | 72.3 | 58.9 | 73.2 |
| PRBench-Legal | 52.7 | 57.6 | 57.6 | 48.5 | 57.6 |
| PRBench-Finance | 51.9 | 55.8 | 55.5 | 46.8 | 58.3 |
| MRCR v2 256K (8-needle) | 83.2 | -- | 93.8 | 86.7 | 92.9 |
| LongBench v2 | 69.1 | -- | 67.1 | 65.3 | 66.3 |
| Opus 4.8 | Fable 5 | GPT 5.6 Sol (max) | Qwen3.7-Max | Qwen3.8-Max | |
|---|---|---|---|---|---|
| 编码代理 | |||||
| Terminal Bench 2.1 | 84.6 | 84.6 | 88.8 | 74.5 | 86.6 |
| SWE-bench Pro | 69.2 | 80.0 | 64.6 | 60.6 | 67.7 |
| DeepSWE 1.1 | 59.0 | 70.0 | 73.0 | 21.6 | 56.6 |
| NL2Repo-Bench | 69.4 | -- | -- | 47.2 | 55.9 |
| FrontierSWE | 70.0 | 88.8 | -- | 40.7 | 73.5 |
| MLS-Bench-Lite | 42.8 | 49.9 | 46.2 | 31.7 | 41.0 |
| PaperBench | 80.3 | 88.8 | 90.5 | 64.8 | 93.0 |
| AndroidBench | 69.8 | 84.5 | 74.0 | 56.5 | 75.1 |
| QwenSWEBench | 84.0 | 86.3 | 73.5 | 63.4 | 80.7 |
| QwenQoderBench | 62.7 | 63.1 | 53.8 | 36.8 | 58.4 |
| QwenReactBench | 1694 | 1770 | 1564 | 1538 | 1724 |
| QwenSVGBench | 1648 | 1690 | 1758 | 1499 | 1713 |
| 通用代理 | |||||
| CoWorkBench | 72.3 | 75.9 | 71.5 | 64.6 | 74.8 |
| WorkSpaceBench | 66.8 | 68.7 | 65.6 | 61.4 | 67.7 |
| JobBench | 48.4 | 57.4 | 45.4 | 31.3 | 53.4 |
| SkillsBench | 65.1 | 70.9 | 73.5 | 61.2 | 70.2 |
| Agents' Last Exam (Pass / Score) | 27.0 / 45.1 | -- / -- | 30.6 / 53.6 | 11.8 / 31.1 | 27.0 / 52.4 |
| Automation-Bench (Pass@1) | 27.2 | 29.1 | 29.7 | 14.2 | 27.3 |
| Toolathlon Verified (Pass@1) | 76.2 | 77.9 | 74.9 | 49.7 | 72.5 |
| WideSearch | 72.9 | 81.2 | -- | 75.2 | 81.9 |
| HLE w/ tools | 57.9 | 64.5 | 58.0 | 53.5 | 56.2 |
| 通用能力 | |||||
| GPQA Diamond | 92.0 | 92.6 | 94.1 | 92.4 | 92.6 |
| HLE | 45.7 | 53.3 | 47.2 | 41.4 | 43.6 |
| IFBench | 62.2 | 63.5 | 72.7 | 79.1 | 82.8 |
| $OneMillion-Bench (expert score) | 41.8 | 55.9 | 53.8 | 44.4 | 52.5 |
| HealthBench | 52.4 | -- | 55.3 | 54.5 | 60.2 |
| PLawBench | 69.6 | 70.2 | 72.3 | 58.9 | 73.2 |
| PRBench-Legal | 52.7 | 57.6 | 57.6 | 48.5 | 57.6 |
| PRBench-Finance | 51.9 | 55.8 | 55.5 | 46.8 | 58.3 |
| MRCR v2 256K (8-needle) | 83.2 | -- | 93.8 | 86.7 | 92.9 |
| LongBench v2 | 69.1 | -- | 67.1 | 65.3 | 66.3 |
1. Fable5 results may involve fallbacks.
1. Fable5 的结果可能涉及回退。
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/).
2. Terminal Bench 2.1：使用 Claude Code 评估（avg@10），超时时间 5 小时，max_tokens=131,072。对于所有其他模型，我们报告各测试框架的最佳公开分数：Claude Opus 4.8 和 Claude Fable 5 使用 Artificial Analysis 的 Terminus 2（https://artificialanalysis.ai/evaluations/terminalbench-v2-1）；GPT-5.6 Sol 使用 Codex（https://openai.com/index/previewing-gpt-5-6-sol/）。
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.
3. SWE-bench Pro：使用 Claude Code 测试框架评估，temp=1.0，top_p=0.95，上下文窗口 256K。修正了有问题的任务，并在精炼后的基准上评估所有基线。
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.
4. DeepSWE 1.1：使用 Claude Code 和 mini-SWE-agent 测试框架评估，temp=1.0，top_p=0.95，上下文窗口 256K。我们报告两个框架中的最高分；值得注意的是，Qwen3.8-Max 在 Claude Code 上表现最佳。
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.
5. NL2Repo-Bench：使用 Claude Code 测试框架评估。为防止奖励黑客，我们禁用尝试访问特定仓库的 Bash 命令，例如 pip download、pip install 和 git clone。
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. "--" indicates that no official MEAN@5 result was available as of that date.
6. FrontierSWE：使用 Claude Code 测试框架评估。所有其他可用的 MEAN@5 结果取自截至 2026 年 8 月 3 日的官方 FrontierSWE 排行榜（https://www.frontierswe.com）。优势分数使用官方评估脚本从原始分数重新计算。"--" 表示截至该日期没有可用的官方 MEAN@5 结果。
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.
7. MLS-Bench-Lite：使用 Claude Code 评估，超时时间 5 小时，max_tokens=131,072。所有其他模型分数取自官方排行榜。
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).
8. PaperBench：在 Code-Dev 模式下的 BasicAgent 设置中评估，由 Claude Opus 4.6 评判，并对 3 次运行取平均（每次运行最长 12 小时）。
9. AndroidBench: Evaluated on the 95-task public subset, reporting avg@3 scores.
9. AndroidBench：在 95 个任务的公共子集上评估，报告 avg@3 分数。
10. QwenSWEBench: Inhouse coding benchmark to evaluate models' 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.
10. QwenSWEBench：内部编码基准，用于评估模型的软件工程能力。使用 Claude Code 测试框架评估。报告 avg@3，超时时间 8 小时，max_tokens=32,768，温度=1.0，上下文窗口 256K 个 token。
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.
11. QwenQoderBench：内部编码基准，用于评估 Qoder 上的用户体验。使用 Claude Code 测试框架评估。报告 avg@5，超时时间 6 小时，max_tokens=32,768，温度=1.0，上下文窗口 256K 个 token。
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.
12. QwenReactBench：内部 React 项目构建基准，使用 Claude Code 作为测试框架，双语（英文/中文），7 个类别；自动渲染 + 多模态评判；BT/Elo 评分。
13. QwenSVGBench: Inhouse SVG code generation benchmark; bilingual (EN/CN), auto-render + multimodal judge; BT/Elo rating.
13. QwenSVGBench：内部SVG代码生成基准；双语（英/中），自动渲染+多模态评判；BT/Elo评分。
14. CoWorkBench: Inhouse cowork benchmark for evaluating long-horizon tasks across computer science, finance, law, medical, and other productivity domains.
14. CoWorkBench：内部协作基准，用于评估计算机科学、金融、法律、医疗及其他生产力领域的长期任务。
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.
15. SkillsBench：在公开的SkillsBench v1.1基准上评估，涵盖87个任务，报告每个任务三次运行的平均得分。Opus 4.8和Fable 5在Claude Code上评估；GPT-5.6 Sol在Codex上评估；Qwen系列在OpenCode上评估。所有结果均来自我们自己的测试。
16. Automation-Bench: Evaluated on the 600-task public subset.
16. Automation-Bench：在600个任务的公开子集上评估。
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.
17. WideSearch：外部模型使用Claude Code测试框架，我们自己的模型使用Qwen-Agent测试框架，报告四次运行的平均item-F1。
18. $OneMillion-Bench: Evaluated using gemini-3.1-pro-preview.
18. $OneMillion-Bench：使用gemini-3.1-pro-preview评估。
19. PLawBench: Evaluated using gemini-3.1-pro-preview.
19. PLawBench：使用gemini-3.1-pro-preview评估。
20. Empty cells (--): Scores are not yet available or are not applicable.
20. 空单元格（--）：分数尚不可用或不适用。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力