Qwen3.6-35B-A3B：智能体编程能力，现已全面开放 | Qwen
Qwen3.6-35B-A3B：智能体编程能力，现已全面开放
QWEN STUDIOHUGGING FACEMODELSCOPEDISCORD
继 Qwen3.6-Plus 发布之后，我们很高兴地开源 Qwen3.6-35B-A3B——一个稀疏但能力卓越的混合专家（MoE）模型，总参数量为 350 亿，仅 30 亿参数被激活。尽管效率极高，Qwen3.6-35B-A3B 仍能提供出色的智能体编程性能，大幅超越其前代 Qwen3.5-35B-A3B，并可媲美 Qwen3.5-27B 和 Gemma4-31B 等规模更大的稠密模型。Qwen3.6-35B-A3B 同时支持多模态思考模式与非思考模式，是当前功能最全面的开源模型之一。现在，Qwen3.6-35B-A3B 已在 Qwen Studio 上线，可通过我们的 API 调用，并以开放权重形式发布给社区。
Qwen3.6-35B-A3B 是一个完全开源的 MoE 模型（总参数量 350 亿 / 激活参数 30 亿），具备以下特点：卓越的智能体编程能力，可与规模大得多的模型相抗衡；强大的多模态感知与推理能力。
您可以在 Qwen Studio 上进行交互式对话，通过阿里云模型服务平台 API 以 Qwen3.6-Flash 名称调用 API，或从 Hugging Face 和 ModelScope 下载权重。
性能表现#
下面我们展示 Qwen3.6-35B-A3B 与同规模模型在广泛任务和模态上的全面评估结果。
语言能力#
仅凭 30 亿激活参数，Qwen3.6-35B-A3B 在多个关键编程基准测试上超越了稠密型 270 亿参数模型 Qwen3.5-27B，并大幅领先其直接前代 Qwen3.5-35B-A3B，尤其是在智能体编程和推理任务上。
| Qwen3.5-27B | Gemma4-31B | Qwen3.5-35BA3B | Gemma4-26BA4B | Qwen3.6-35BA3B | |
|---|---|---|---|---|---|
| 编程智能体 | |||||
| SWE-bench Verified | 75.0 | 52.0 | 70.0 | 17.4 | 73.4 |
| SWE-bench Multilingual | 69.3 | 51.7 | 60.3 | 17.3 | 67.2 |
| SWE-bench Pro | 51.2 | 35.7 | 44.6 | 13.8 | 49.5 |
| Terminal-Bench 2.0 | 41.6 | 42.9 | 40.5 | 34.2 | 51.5 |
| Claw-Eval 平均分 | 64.3 | 48.5 | 65.4 | 58.8 | 68.7 |
| Claw-Eval Pass^3 | 46.2 | 25.0 | 51.0 | 28.0 | 50.0 |
| SkillsBench 平均分5 | 27.2 | 23.6 | 4.4 | 12.3 | 28.7 |
| QwenClawBench | 52.2 | 41.7 | 47.7 | 38.7 | 52.6 |
| NL2Repo | 27.3 | 15.5 | 20.5 | 11.6 | 29.4 |
| QwenWebBench | 1068 | 1197 | 978 | 1178 | 1397 |
| 通用智能体 | |||||
| TAU3-Bench | 68.4 | 67.5 | 68.9 | 59.0 | 67.2 |
| VITA-Bench | 41.8 | 43.0 | 29.1 | 36.9 | 35.6 |
| DeepPlanning | 22.6 | 24.0 | 22.8 | 16.2 | 25.9 |
| Tool Decathlon | 31.5 | 21.2 | 28.7 | 12.0 | 26.9 |
| MCPMark | 36.3 | 18.1 | 27.0 | 14.2 | 37.0 |
| MCP-Atlas | 68.4 | 57.2 | 62.4 | 50.0 | 62.8 |
| WideSearch | 66.4 | 35.2 | 59.1 | 38.3 | 60.1 |
| 知识 | |||||
| MMLU-Pro | 86.1 | 85.2 | 85.3 | 82.6 | 85.2 |
| MMLU-Redux | 93.2 | 93.7 | 93.3 | 92.7 | 93.3 |
| SuperGPQA | 65.6 | 65.7 | 63.4 | 61.4 | 64.7 |
| C-Eval | 90.5 | 82.6 | 90.2 | 82.5 | 90.0 |
| STEM 与推理 | |||||
| GPQA | 85.5 | 84.3 | 84.2 | 82.3 | 86.0 |
| HLE | 24.3 | 19.5 | 22.4 | 8.7 | 21.4 |
| LiveCodeBench v6 | 80.7 | 80.0 | 74.6 | 77.1 | 80.4 |
| HMMT 2025年2月 | 92.0 | 88.7 | 89.0 | 91.7 | 90.7 |
| HMMT 2025年11月 | 89.8 | 87.5 | 89.2 | 87.5 | 89.1 |
| HMMT 2026年2月 | 84.3 | 77.2 | 78.7 | 79.0 | 83.6 |
| IMOAnswerBench | 79.9 | 74.5 | 76.8 | 74.3 | 78.9 |
| AIME26 | 92.6 | 89.2 | 91.0 | 88.3 | 92.7 |
SWE-Bench 系列：内部智能体脚手架（bash + 文件编辑工具）；temp=1.0，topp=0.95，200K 上下文窗口。我们修正了 SWE-bench Pro 公开数据集中的一些有问题的任务，并在优化后的基准上评估所有基线。
Terminal-Bench 2.0：Harbor/Terminus-2 测试框架；3 小时超时，32 CPU/48 GB 内存；temp=1.0，topp=0.95，topk=20，maxtokens=80K，256K 上下文；5 次运行的平均值。
SkillsBench：通过 OpenCode 在 78 个任务上评估（自包含子集，排除依赖 API 的任务）；5 次运行的平均值。
NL2Repo：其他模型通过 Claude Code 评估（temp=1.0，topp=0.95，maxturns=900）。
QwenClawBench：一个基于真实用户分布的内部 Claw 智能体基准（即将开源）；temp=0.6，256K 上下文。
QwenWebBench：一个内部前端代码生成基准；双语（英文/中文），7 个类别（网页设计、网页应用、游戏、SVG、数据可视化、动画和 3D）；自动渲染 + 多模态评判器（代码/视觉正确性）；BT/Elo 评分系统。
TAU3-Bench：我们使用官方用户模型（gpt-5.2，低推理强度）+ 默认 BM25 检索。
VITA-Bench：各子领域平均得分；使用 claude-4-sonnet 作为评判器，因为官方评判器（claude-3.7-sonnet）已不再可用。
MCPMark：GitHub MCP v0.30.3；Playwright 响应截断至 32K tokens。
MCP-Atlas：公开集得分；gemini-2.5-pro 评判器。
AIME 26：我们使用完整的 AIME 2026（I 和 II），得分可能与 Qwen 3.5 说明中的结果不同。
视觉语言
Qwen3.6 原生支持多模态，Qwen3.6-35B-A3B 展现出的感知与多模态推理能力远超其模型规模所暗示的水平，仅需约 30 亿激活参数。在大多数视觉语言基准测试中，其性能与 Claude Sonnet 4.5 相当，甚至在多项任务上超越后者。该模型的优势在空间智能方面尤为突出，在 RefCOCO 上达到 92.0，在 ODInW13 上达到 50.8。
| Qwen3.5-27B | Claude-Sonnet-4.5 | Gemma4-31B | Gemma4-26BA4B | Qwen3.5-35B-A3B | Qwen3.6-35B-A3B | |
|---|---|---|---|---|---|---|
| STEM 与谜题 | ||||||
| MMMU | 82.3 | 79.6 | 80.4 | 78.4 | 81.4 | 81.7 |
| MMMU-Pro | 75.0 | 68.4 | 76.9 | 73.8 | 75.1 | 75.3 |
| Mathvista(mini) | 87.8 | 79.8 | 79.3 | 79.4 | 86.2 | 86.4 |
| ZEROBenchsub | 36.2 | 26.3 | 26.0 | 26.3 | 34.1 | 34.4 |
| 通用 VQA | ||||||
| RealWorldQA | 83.7 | 70.3 | 72.3 | 72.2 | 84.1 | 85.3 |
| MMBench EN-DEV-v1.1 | 92.6 | 88.3 | 90.9 | 89.0 | 91.5 | 92.8 |
| SimpleVQA | 56.0 | 57.6 | 52.9 | 52.2 | 58.3 | 58.9 |
| HallusionBench | 70.0 | 59.9 | 67.4 | 66.1 | 67.9 | 69.8 |
| 文本识别与文档理解 | ||||||
| OmniDocBench1.5 | 88.9 | 85.8 | 80.1 | 74.4 | 89.3 | 89.9 |
| CharXiv(RQ) | 79.5 | 67.2 | 67.9 | 69.0 | 77.5 | 78.0 |
| CC-OCR | 81.0 | 68.1 | 75.7 | 74.5 | 80.7 | 81.9 |
| AI2DTEST | 92.9 | 87.0 | 89.0 | 88.3 | 92.6 | 92.7 |
| 空间智能 | ||||||
| RefCOCO(avg) | 90.9 | -- | -- | -- | 89.2 | 92.0 |
| ODInW13 | 41.1 | -- | -- | -- | 42.6 | 50.8 |
| EmbSpatialBench | 84.5 | 71.8 | -- | -- | 83.1 | 84.3 |
| RefSpatialBench | 67.7 | -- | -- | -- | 63.5 | 64.3 |
| 视频理解 | ||||||
| VideoMME(w sub.) | 87.0 | 81.1 | -- | -- | 86.6 | 86.6 |
| VideoMME(w/o sub.) | 82.8 | 75.3 | -- | -- | 82.5 | 82.5 |
| VideoMMMU | 82.3 | 77.6 | 81.6 | 76.0 | 80.4 | 83.7 |
| MLVU | 85.9 | 72.8 | -- | -- | 85.6 | 86.2 |
| MVBench | 74.6 | -- | -- | -- | 74.8 | 74.6 |
| LVBench | 73.6 | -- | -- | -- | 71.4 | 71.4 |
空单元格（--）表示分数不可用或不适用。
使用 Qwen3.6-35B-A3B 构建#
Qwen3.6-35B-A3B 以开放权重形式在 Hugging Face 和 ModelScope 上提供，可用于自行部署，也可通过阿里云百炼 API 以 qwen3.6-flash 名称调用。您还可以在 Qwen Studio 上立即体验。
该模型可与流行的第三方编程助手无缝集成，包括 OpenClaw、Claude Code 和 Qwen Code，从而简化开发工作流程，实现高效且具有上下文感知能力的编程体验。
API 使用#
本次发布支持保留思考功能：保留消息中所有先前轮次的思考内容，建议在智能体任务中使用。
阿里云百炼
阿里云百炼平台支持行业标准协议，包括兼容 OpenAI 规范的聊天补全与响应 API，以及兼容 Anthropic 的 API 接口。
以下是聊天补全 API 的示例代码：
""" 环境变量（根据官方文档）： DASHSCOPEAPIKEY：您的 API 密钥，来自 https://modelstudio.console.alibabacloud.com DASHSCOPEBASEURL：（可选）兼容模式 API 的基础 URL。
- 北京：https://dashscope.aliyuncs.com/compatible-mode/v1
- 新加坡：https://dashscope-intl.aliyuncs.com/compatible-mode/v1
- 美国（弗吉尼亚）：https://dashscope-us.aliyuncs.com/compatible-mode/v1 DASHSCOPEMODEL：（可选）模型名称；可覆盖以使用不同模型。 """from openai import OpenAIimport osapikey = os.environ.get("DASHSCOPEAPIKEY")if not apikey: raise ValueError( "DASHSCOPEAPIKEY is required. " "Set it via: export DASHSCOPEAPIKEY='your-api-key'" )client = OpenAI( apikey=apikey, baseurl=os.environ.get( "DASHSCOPEBASEURL", "https://dashscope-intl.aliyuncs.com/compatible-mode/v1", ),)messages = [{"role": "user", "content": "介绍 vibe coding。"}]model = os.environ.get( "DASHSCOPEMODEL", "qwen3.6-flash",)completion = client.chat.completions.create( model=model, messages=messages, extrabody={ "enablethinking": True, # "preservethinking": True, }, stream=True)reasoningcontent = "" # 完整推理过程answercontent = "" # 完整回答isanswering = False # 是否已进入回答阶段print("\n" + "=" 20 + "Reasoning" + "=" 20 + "\n")for chunk in completion: if not chunk.choices: print("\nUsage:") print(chunk.usage) continue delta = chunk.choices[0].delta # 仅收集推理内容 if hasattr(delta, "reasoningcontent") and delta.reasoningcontent is not None: if not isanswering: print(delta.reasoningcontent, end="", flush=True) reasoningcontent += delta.reasoningcontent # 收到内容，开始回答阶段 if hasattr(delta, "content") and delta.content: if not isanswering: print("\n" + "=" 20 + "Answer" + "=" 20 + "\n") isanswering = True print(delta.content, end="", flush=True) answercontent += delta.content
更多信息，请访问 API 文档。
编程与智能体
Qwen3.6-35B-A3B 具备出色的智能体编码能力，可无缝集成到流行的第三方编码助手，包括 OpenClaw、Claude Code 和 Qwen Code。
OpenClaw#
Qwen3.6-35B-A3B 兼容 OpenClaw（原名 Moltbot / Clawdbot），这是一款自托管开源 AI 编码智能体。将其连接到 Model Studio，即可在终端中获得完整的智能体编码体验。通过以下脚本开始使用：
bash
Node.js 22+curl -fsSL https://molt.bot/install.sh | bash # macOS / Linux# 设置你的 API 密钥export DASHSCOPEAPIKEY= 启动 OpenClawopenclaw dashboard # 网页浏览器# openclaw tui # 打开新终端并启动 TUI
首次使用时，编辑 ~/.openclaw/openclaw.json 文件，将 OpenClaw 指向 Model Studio。找到或创建以下字段并进行合并——不要覆盖整个文件，以保留你现有的设置：
json
{ "models": { "mode": "merge", "providers": { "modelstudio": { "baseUrl": "https://dashscope-intl.aliyuncs.com/compatible-mode/v1", "apiKey": "DASHSCOPEAPIKEY", "api": "openai-completions", "models": [ { "id": "qwen3.6-flash", "name": "qwen3.6-flash", "reasoning": true, "input": ["text", "image"], "contextWindow": 131072, "maxTokens": 16384 } ] } } }, "agents": { "defaults": { "model": { "primary": "modelstudio/qwen3.6-flash" }, "models": { "modelstudio/qwen3.6-flash": {} } } }}
Qwen Code#
Qwen3.6-35B-A3B 兼容 Qwen Code，这是一款专为终端设计、并针对通义千问系列深度优化的开源 AI 智能体。通过以下脚本开始使用：
bash
Node.js 20+npm install -g @qwen-code/qwen-code@latest 启动 Qwen Code（交互模式）qwen 然后在会话中：/help /auth
首次使用时，系统会提示你登录。你可以随时运行 /auth 来切换认证方式。
Claude Code#
Qwen API 还支持 Anthropic API 协议，这意味着你可以将其与 Claude Code 等工具配合使用，以获得更高级的编码体验：
bash
安装 Claude Codenpm install -g @anthropic-ai/claude-code 配置环境变量export ANTHROPICMODEL="qwen3.6-flash"export ANTHROPICSMALLFASTMODEL="qwen3.6-flash"export ANTHROPICBASEURL=https://dashscope-intl.aliyuncs.com/apps/anthropic export ANTHROPICAUTHTOKEN= 启动 CLIclaude
总结
Qwen3.6-35B-A3B 证明了稀疏 MoE 模型能够实现卓越的智能体编码与推理能力。仅需 3B 活跃参数，其性能即可媲美活跃规模数倍于己的稠密模型，同时在多模态基准测试中也表现优异。作为一款完全开源模型，它为其规模层级所能达到的水平树立了新标杆。
展望未来，我们将持续扩展 Qwen3.6 开源家族，并不断突破高效开源模型的能力边界。我们感谢社区的反馈，并期待看到大家基于 Qwen3.6-35B-A3B 构建的成果。此外，Qwen3.6 开源家族仍在持续壮大，敬请关注我们后续的发布！
引用说明#
如果您认为 Qwen3.6-35B-A3B 对您有帮助，欢迎引用以下文章：
@misc{qwen3635ba3b, title = {{Qwen3.6-35B-A3B}: 智能体编码能力，现已全面开放}, url = {https://qwen.ai/blog?id=qwen3.6-35b-a3b}, author = {{Qwen 团队}}, month = {四月}, year = {2026}}
Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All | Qwen
Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All
QWEN STUDIOHUGGING FACEMODELSCOPEDISCORD
Following the launch of Qwen3.6-Plus, we are excited to open-source Qwen3.6-35B-A3B — a sparse yet remarkably capable mixture-of-experts (MoE) model with 35 billion total parameters and only 3 billion active parameters. Despite its efficiency, Qwen3.6-35B-A3B delivers outstanding agentic coding performance, surpassing its predecessor Qwen3.5-35B-A3B by a wide margin and rivaling much larger dense models such as Qwen3.5-27B and Gemma4-31B. Still supporting both multimodal thinking and non-thinking modes, Qwen3.6-35B-A3B works as one of the most versatile open-source models available today. Now, Qwen3.6-35B-A3B is live on Qwen Studio, available through our API, and released as open weights for the community.
Qwen3.6-35B-A3B is a fully open-source MoE model (35B total / 3B active), featuring: exceptional agentic coding capability competitive with much larger models strong multimodal perception and reasoning ability
You can chat interactively on Qwen Studio, call via API as Qwen3.6-Flash on Alibaba Cloud Model Studio API, or download weights from Hugging Face and ModelScope.
Performance#
Below we present comprehensive evaluations of Qwen3.6-35B-A3B against peer-scale models across a wide range of tasks and modalities.
Language#
With only 3B active parameters, Qwen3.6-35B-A3B outperforms the dense 27B-parameter Qwen3.5-27B on several key coding benchmarks and dramatically surpasses its direct predecessor Qwen3.5-35B-A3B, especially on agentic coding and reasoning tasks.
| Qwen3.5-27B | Gemma4-31B | Qwen3.5-35BA3B | Gemma4-26BA4B | Qwen3.6-35BA3B | |
|---|---|---|---|---|---|
| Coding Agent | |||||
| SWE-bench Verified | 75.0 | 52.0 | 70.0 | 17.4 | 73.4 |
| SWE-bench Multilingual | 69.3 | 51.7 | 60.3 | 17.3 | 67.2 |
| SWE-bench Pro | 51.2 | 35.7 | 44.6 | 13.8 | 49.5 |
| Terminal-Bench 2.0 | 41.6 | 42.9 | 40.5 | 34.2 | 51.5 |
| Claw-Eval Avg | 64.3 | 48.5 | 65.4 | 58.8 | 68.7 |
| Claw-Eval Pass^3 | 46.2 | 25.0 | 51.0 | 28.0 | 50.0 |
| SkillsBench Avg5 | 27.2 | 23.6 | 4.4 | 12.3 | 28.7 |
| QwenClawBench | 52.2 | 41.7 | 47.7 | 38.7 | 52.6 |
| NL2Repo | 27.3 | 15.5 | 20.5 | 11.6 | 29.4 |
| QwenWebBench | 1068 | 1197 | 978 | 1178 | 1397 |
| General Agent | |||||
| TAU3-Bench | 68.4 | 67.5 | 68.9 | 59.0 | 67.2 |
| VITA-Bench | 41.8 | 43.0 | 29.1 | 36.9 | 35.6 |
| DeepPlanning | 22.6 | 24.0 | 22.8 | 16.2 | 25.9 |
| Tool Decathlon | 31.5 | 21.2 | 28.7 | 12.0 | 26.9 |
| MCPMark | 36.3 | 18.1 | 27.0 | 14.2 | 37.0 |
| MCP-Atlas | 68.4 | 57.2 | 62.4 | 50.0 | 62.8 |
| WideSearch | 66.4 | 35.2 | 59.1 | 38.3 | 60.1 |
| Knowledge | |||||
| MMLU-Pro | 86.1 | 85.2 | 85.3 | 82.6 | 85.2 |
| MMLU-Redux | 93.2 | 93.7 | 93.3 | 92.7 | 93.3 |
| SuperGPQA | 65.6 | 65.7 | 63.4 | 61.4 | 64.7 |
| C-Eval | 90.5 | 82.6 | 90.2 | 82.5 | 90.0 |
| STEM & Reasoning | |||||
| GPQA | 85.5 | 84.3 | 84.2 | 82.3 | 86.0 |
| HLE | 24.3 | 19.5 | 22.4 | 8.7 | 21.4 |
| LiveCodeBench v6 | 80.7 | 80.0 | 74.6 | 77.1 | 80.4 |
| HMMT Feb 25 | 92.0 | 88.7 | 89.0 | 91.7 | 90.7 |
| HMMT Nov 25 | 89.8 | 87.5 | 89.2 | 87.5 | 89.1 |
| HMMT Feb 26 | 84.3 | 77.2 | 78.7 | 79.0 | 83.6 |
| IMOAnswerBench | 79.9 | 74.5 | 76.8 | 74.3 | 78.9 |
| AIME26 | 92.6 | 89.2 | 91.0 | 88.3 | 92.7 |
SWE-Bench Series: Internal agent scaffold (bash + file-edit tools); temp=1.0, topp=0.95, 200K context window. We correct some problematic tasks in the public set of SWE-bench Pro and evaluate all baselines on the refined benchmark.
Terminal-Bench 2.0: Harbor/Terminus-2 harness; 3h timeout, 32 CPU/48 GB RAM; temp=1.0, topp=0.95, topk=20, maxtokens=80K, 256K ctx; avg of 5 runs.
SkillsBench: Evaluated via OpenCode on 78 tasks (self-contained subset, excluding API-dependent tasks); avg of 5 runs.
NL2Repo: Others are evaluated via Claude Code (temp=1.0, topp=0.95, maxturns=900).
QwenClawBench: An internal real-user-distribution Claw agent benchmark (open-sourcing soon); temp=0.6, 256K ctx.
QwenWebBench: An internal front-end code generation benchmark; bilingual (EN/CN), 7 categories (Web Design, Web Apps, Games, SVG, Data Visualization, Animation, and 3D); auto-render + multimodal judge (code/visual correctness); BT/Elo rating system.
TAU3-Bench: We use the official user model (gpt-5.2, low reasoning effort) + default BM25 retrieval.
VITA-Bench: Avg subdomain scores; using claude-4-sonnet as judger, as the official judger (claude-3.7-sonnet) is no longer available.
MCPMark: GitHub MCP v0.30.3; Playwright responses truncated at 32K tokens.
MCP-Atlas: Public set score; gemini-2.5-pro judger.
AIME 26: We use the full AIME 2026 (I & II), where the scores may differ from Qwen 3.5 notes.
Vision Language#
Qwen3.6 is natively multimodal, and Qwen3.6-35B-A3B showcases perception and multimodal reasoning capabilities that far exceed what its size would suggest, with only around 3 billion activated parameters. Across most vision-language benchmarks, its performance matches Claude Sonnet 4.5, and even surpasses it on several tasks. Its strengths are particularly evident in spatial intelligence, where it achieves 92.0 on RefCOCO and 50.8 on ODInW13.
| Qwen3.5-27B | Claude-Sonnet-4.5 | Gemma4-31B | Gemma4-26BA4B | Qwen3.5-35B-A3B | Qwen3.6-35B-A3B | |
|---|---|---|---|---|---|---|
| STEM and Puzzle | ||||||
| MMMU | 82.3 | 79.6 | 80.4 | 78.4 | 81.4 | 81.7 |
| MMMU-Pro | 75.0 | 68.4 | 76.9 | 73.8 | 75.1 | 75.3 |
| Mathvista(mini) | 87.8 | 79.8 | 79.3 | 79.4 | 86.2 | 86.4 |
| ZEROBenchsub | 36.2 | 26.3 | 26.0 | 26.3 | 34.1 | 34.4 |
| General VQA | ||||||
| RealWorldQA | 83.7 | 70.3 | 72.3 | 72.2 | 84.1 | 85.3 |
| MMBench EN-DEV-v1.1 | 92.6 | 88.3 | 90.9 | 89.0 | 91.5 | 92.8 |
| SimpleVQA | 56.0 | 57.6 | 52.9 | 52.2 | 58.3 | 58.9 |
| HallusionBench | 70.0 | 59.9 | 67.4 | 66.1 | 67.9 | 69.8 |
| Text Recognition and Document Understanding | ||||||
| OmniDocBench1.5 | 88.9 | 85.8 | 80.1 | 74.4 | 89.3 | 89.9 |
| CharXiv(RQ) | 79.5 | 67.2 | 67.9 | 69.0 | 77.5 | 78.0 |
| CC-OCR | 81.0 | 68.1 | 75.7 | 74.5 | 80.7 | 81.9 |
| AI2DTEST | 92.9 | 87.0 | 89.0 | 88.3 | 92.6 | 92.7 |
| Spatial Intelligence | ||||||
| RefCOCO(avg) | 90.9 | -- | -- | -- | 89.2 | 92.0 |
| ODInW13 | 41.1 | -- | -- | -- | 42.6 | 50.8 |
| EmbSpatialBench | 84.5 | 71.8 | -- | -- | 83.1 | 84.3 |
| RefSpatialBench | 67.7 | -- | -- | -- | 63.5 | 64.3 |
| Video Understanding | ||||||
| VideoMME(w sub.) | 87.0 | 81.1 | -- | -- | 86.6 | 86.6 |
| VideoMME(w/o sub.) | 82.8 | 75.3 | -- | -- | 82.5 | 82.5 |
| VideoMMMU | 82.3 | 77.6 | 81.6 | 76.0 | 80.4 | 83.7 |
| MLVU | 85.9 | 72.8 | -- | -- | 85.6 | 86.2 |
| MVBench | 74.6 | -- | -- | -- | 74.8 | 74.6 |
| LVBench | 73.6 | -- | -- | -- | 71.4 | 71.4 |
Empty cells (--) indicate scores not available or not applicable.
Build with Qwen3.6-35B-A3B#
Qwen3.6-35B-A3B is available as open weights on Hugging Face and ModelScope for self-hosting, and through the Alibaba Cloud Model Studio API as qwen3.6-flash. You can also try it instantly on Qwen Studio.
The model can be seamlessly integrated with popular third-party coding assistants, including OpenClaw, Claude Code, and Qwen Code, to streamline development workflows and enable efficient, context-aware coding experiences.
API Usage#
This release supports the preservethinking feature: preserving thinking content from all preceding turns in messages, which is recommended for agentic tasks.
Alibaba Cloud Model Studio#
Alibaba Cloud Model Studio supports industry-standard protocols, including chat completions and responses APIs compatible with OpenAI’s specification, as well as an API interface compatible with Anthropic.
Example code for chat completions API is provided below:
""" Environment variables (per official docs): DASHSCOPEAPIKEY: Your API Key from https://modelstudio.console.alibabacloud.com DASHSCOPEBASEURL: (optional) Base URL for compatible-mode API.
- Beijing: https://dashscope.aliyuncs.com/compatible-mode/v1
- Singapore: https://dashscope-intl.aliyuncs.com/compatible-mode/v1
- US (Virginia): https://dashscope-us.aliyuncs.com/compatible-mode/v1 DASHSCOPEMODEL: (optional) Model name; override for different models. """from openai import OpenAIimport osapikey = os.environ.get("DASHSCOPEAPIKEY")if not apikey: raise ValueError( "DASHSCOPEAPIKEY is required. " "Set it via: export DASHSCOPEAPIKEY='your-api-key'" )client = OpenAI( apikey=apikey, baseurl=os.environ.get( "DASHSCOPEBASEURL", "https://dashscope-intl.aliyuncs.com/compatible-mode/v1", ),)messages = [{"role": "user", "content": "Introduce vibe coding."}]model = os.environ.get( "DASHSCOPEMODEL", "qwen3.6-flash",)completion = client.chat.completions.create( model=model, messages=messages, extrabody={ "enablethinking": True, # "preservethinking": True, }, stream=True)reasoningcontent = "" # Full reasoning traceanswercontent = "" # Full responseisanswering = False # Whether we have entered the answer phaseprint("\n" + "=" 20 + "Reasoning" + "=" 20 + "\n")for chunk in completion: if not chunk.choices: print("\nUsage:") print(chunk.usage) continue delta = chunk.choices[0].delta # Collect reasoning content only if hasattr(delta, "reasoningcontent") and delta.reasoningcontent is not None: if not isanswering: print(delta.reasoningcontent, end="", flush=True) reasoningcontent += delta.reasoningcontent # Received content, start answer phase if hasattr(delta, "content") and delta.content: if not isanswering: print("\n" + "=" 20 + "Answer" + "=" 20 + "\n") isanswering = True print(delta.content, end="", flush=True) answercontent += delta.content
For more information, please visit the API doc.
Coding & Agents#
Qwen3.6-35B-A3B features excellent agentic coding capabilities and can be seamlessly integrated into popular third-party coding assistants, including OpenClaw, Claude Code, and Qwen Code.
OpenClaw#
Qwen3.6-35B-A3B is compatible with OpenClaw (formerly Moltbot / Clawdbot), a self-hosted open-source AI coding agent. Connect it to Model Studio to get a full agentic coding experience in the terminal. Get started with the following script:
bash
Node.js 22+curl -fsSL https://molt.bot/install.sh | bash # macOS / Linux# Set your API keyexport DASHSCOPEAPIKEY= Launch OpenClawopenclaw dashboard # web browser# openclaw tui # Open a new terminal and start the TUI
On first use, edit ~/.openclaw/openclaw.json to point OpenClaw at Model Studio. Find or create the following fields and merge them — do not overwrite the entire file to preserve your existing settings:
json
{ "models": { "mode": "merge", "providers": { "modelstudio": { "baseUrl": "https://dashscope-intl.aliyuncs.com/compatible-mode/v1", "apiKey": "DASHSCOPEAPIKEY", "api": "openai-completions", "models": [ { "id": "qwen3.6-flash", "name": "qwen3.6-flash", "reasoning": true, "input": ["text", "image"], "contextWindow": 131072, "maxTokens": 16384 } ] } } }, "agents": { "defaults": { "model": { "primary": "modelstudio/qwen3.6-flash" }, "models": { "modelstudio/qwen3.6-flash": {} } } }}
Qwen Code#
Qwen3.6-35B-A3B is compatible with Qwen Code, an open-source AI agent designed for the terminal and deeply optimized for the Qwen Series. Get started with the following script:
bash
Node.js 20+npm install -g @qwen-code/qwen-code@latest Start Qwen Code (interactive)qwen Then, in the session:/help /auth
On first use, you’ll be prompted to sign in. You can run /auth anytime to switch authentication methods.
Claude Code#
Qwen APIs also support the Anthropic API protocol, meaning you can use it with tools like Claude Code for elevated coding experience:
bash
Install Claude Codenpm install -g @anthropic-ai/claude-code Configure environmentexport ANTHROPICMODEL="qwen3.6-flash"export ANTHROPICSMALLFASTMODEL="qwen3.6-flash"export ANTHROPICBASEURL=https://dashscope-intl.aliyuncs.com/apps/anthropic export ANTHROPICAUTHTOKEN= Launch the CLIclaude
Summary#
Qwen3.6-35B-A3B demonstrates that sparse MoE models can achieve remarkable agentic coding and reasoning capability. With only 3B active parameters, it delivers performance that rivals dense models several times its active size, while also excelling across multimodal benchmarks. As a fully open-source checkpoint, it sets a new standard for what’s possible at its scale.
Looking ahead, we will continue to expand the Qwen3.6 open-source family and push the boundaries of what efficient, open models can accomplish. We are grateful for the community’s feedback and look forward to seeing what you build with Qwen3.6-35B-A3B. Also, Qwen3.6 open-source family keeps expanding, stay tuned for our future releases!
Citation#
Feel free to cite the following article if you find Qwen3.6-35B-A3B helpful:
@misc{qwen3635ba3b, title = {{Qwen3.6-35B-A3B}: Agentic Coding Power, Now Open to All}, url = {https://qwen.ai/blog?id=qwen3.6-35b-a3b}, author = {{Qwen Team}}, month = {April}, year = {2026}}