What is Qwen3.6-Max-Preview
Qwen3.6-Max-Preview is an early preview version of the next-generation flagship large model launched by the Alibaba Tongyi Qianwen team. Compared with Qwen3.6-Plus, the model achieves significant improvements in three major dimensions: agentic coding, world knowledge, and instruction following, and achieves the highest scores in six coding benchmarks including SWE-bench Pro, Terminal-Bench 2.0, and SciCode. Users can experience it online through Qwen Studio or call it through the Alibaba Cloud Model Studio API.
Main features of Qwen3.6-Max-Preview
-
Agentic coding: Achieves the highest scores in six coding benchmarks including SWE-bench Pro, Terminal-Bench 2.0, and SkillsBench, supporting code generation, terminal operations, and long-horizon coding tasks.
-
World knowledge enhancement: Graduate-level knowledge question answering (SuperGPQA) and Chinese knowledge (QwenChineseBench) capabilities are significantly improved.
-
Instruction following optimization: Better performance in real-world tool call format following (ToolcallFormatIFBench).
-
Chain-of-thought retention: Supports
preserve_thinkingfunctionality, which can retain previous thinking content in multi-turn conversations, suitable for complex agentic tasks. -
Protocol compatibility: The API is compatible with OpenAI (chat completions/responses) and Anthropic interface specifications, making it easy to seamlessly integrate into existing workflows.
How to use Qwen3.6-Max-Preview
- Online experience: Visit Qwen Studio, select the Qwen3.6-Max-Preview model to start interactive conversation.
- API call
- VisitAlibaba Cloud Model Studioofficial website to obtain an API Key.
- When calling, enter the model name
qwen3.6-max-preview。 - Supports the OpenAI standard protocol (chat completions / responses) and the Anthropic API interface, allowing direct replacement of the model name in existing workflows for integration.
Key information and usage requirements of Qwen3.6-Max-Preview
-
Product positioning: An early preview version of the next-generation flagship large model from Alibaba Tongyi Qianwen, an upgraded version after Qwen3.6-Plus.
-
Core improvements: Agentic coding capabilities are significantly enhanced, with better performance in world knowledge and instruction following.
-
Performance: Achieves the highest scores in six coding benchmarks: SWE-bench Pro, Terminal-Bench 2.0, SkillsBench, QwenClawBench, QwenWebBench, and SciCode.
-
Status note: It is still a preview version, and the model is actively iterating, with subsequent versions to be continuously optimized.
Core advantages of Qwen3.6-Max-Preview
-
Leading agentic coding: Achieves the highest scores in six coding benchmarks including SWE-bench Pro, Terminal-Bench 2.0, and SkillsBench, with code generation and terminal operation capabilities greatly surpassing the previous generation.
-
World knowledge enhancement: Graduate-level knowledge question answering (SuperGPQA) and Chinese knowledge (QwenChineseBench) performance are significantly improved, with better knowledge reliability.
-
More precise instruction following: Stronger real-world tool call format following (ToolcallFormatIFBench) capability, with more stable execution of complex tasks.
-
Outstanding competitive performance: Outperforms mainstream frontier models such as Claude 4.5 Opus and GLM 5.1 on multiple benchmarks.
-
Strong ecosystem compatibility: The API is compatible with the OpenAI and Anthropic protocols, and supports
preserve_thinkingfeatures, enabling seamless integration into existing agent workflows.
Comparison of Qwen3.6-Max-Preview with similar competitors
| Comparison dimension | Qwen3.6-Max-Preview | Claude 4.5 Opus | GLM 5.1 |
|---|---|---|---|
| SWE-bench Pro(engineering code) | 57.3 🥇 | 50.9 | 58.4 |
| Terminal-Bench 2.0(terminal programming) | 65.4 🥇 | 59.3 | 63.5 |
| SkillsBench(agent skills) | 55.6 🥇 | 45.3 | 53.1 |
| SciCode(scientific research code) | 47.0 | 49.5 🥇 | — |
| NL2Repo(long-horizon code) | 42.9 | 43.2 🥇 | 42.7 |
| QwenWebBench(front-end development) | 1532 | — | 1558 🥇 |
| SuperGPQA(graduate-level knowledge) | 73.9 🥇 | 70.6 | — |
| QwenChineseBench(Chinese knowledge) | 84.0 🥇 | 69.0 | — |
| ToolcallFormatIFBench(instruction following) | 86.1 🥇 | 84.2 | 60.1 |
| GDPval-AA(real-world tasks) | 51.0 🥇 | 48.0 | 52.0 |
Application scenarios of Qwen3.6-Max-Preview
-
Software development: Supports code generation, debugging, terminal operations, and repository-level long-horizon programming tasks, adapting to full-stack development workflows.
-
Scientific computing: Assists with research-oriented code writing and data analysis, improving the efficiency of research automation.
-
Front-end construction: Generates web designs, web applications, games, SVG, data visualizations, animations, and 3D content.
-
Agent automation: The model can perform complex real-world tasks, chaining multi-step workflows through tool calls.
-
Knowledge Q&A: The model provides graduate-level professional knowledge answers and in-depth knowledge services in Chinese domains.