Qwen 3.6 Max Preview
The flagship that tops 6 major coding benchmarks and leads Chinese LLMs
Qwen 3.6 Max Preview is Alibaba's proprietary flagship MoE model, released April 20, 2026. It tops 6 major coding benchmarks, scores 86% on GPQA (PhD-level science), and achieves an AA-Intelligence Index score of 52 - the highest among Chinese LLMs. Features 256K context and preserve_thinking for agentic workflows.
Capabilities
Flagship intelligence across coding, science, and instruction following
Qwen 3.6 Max Preview combines MoE architecture with significant improvements in coding, instruction following, and world knowledge to deliver Alibaba's most capable model to date.
Coding benchmark leader
Tops 6 major coding benchmarks with significant improvements over Qwen 3.6 Plus on SWE-bench Pro. The flagship model for complex software engineering, code generation, and debugging tasks.
PhD-level science
86% on GPQA for graduate and PhD-level scientific reasoning. Demonstrates deep understanding across physics, chemistry, biology, and interdisciplinary scientific domains.
AA-Intelligence Index leader
Score of 52 on the AA-Intelligence Index - the highest among all Chinese LLMs. A comprehensive measure of general intelligence across reasoning, knowledge, and task completion.
Superior instruction following
Significant improvements in instruction following over Qwen 3.6 Plus. Precisely executes complex, multi-constraint instructions with high fidelity and consistency.
Enhanced world knowledge
Major advances in world knowledge breadth and accuracy. Provides well-grounded, factual responses across diverse domains from history and geography to current events and technical topics.
preserve_thinking for agents
The preserve_thinking parameter maintains reasoning state across agent loop iterations. Combined with 256K context, it enables complex multi-step agentic workflows without losing chain-of-thought context.
Key highlights
Alibaba's most capable model
Qwen 3.6 Max Preview achieves top-tier results across coding, science, and general intelligence benchmarks, establishing itself as the flagship of the Qwen 3.6 family.
Top achievements
- Tops 6 major coding benchmarks
- GPQA: 86% - PhD-level scientific reasoning
- AA-Intelligence Index: 52 - highest among Chinese LLMs
- Significant SWE-bench Pro improvements over Plus
- Major advances in instruction following and world knowledge
Technical specs
- Proprietary flagship MoE model by Alibaba Cloud
- 256K token context window
- preserve_thinking parameter for agent loops
- OpenAI-compatible API
- Released April 20, 2026 (Preview)
Performance
Flagship performance across coding, science, and general intelligence
Qwen 3.6 Max Preview tops 6 major coding benchmarks, scores 86% on GPQA, and achieves the highest AA-Intelligence Index score among Chinese LLMs at 52.
Qwen 3.6 Max Preview represents a significant leap over Qwen 3.6 Plus, with improvements across coding, instruction following, world knowledge, and scientific reasoning - establishing itself as Alibaba's most capable model.
Tops 6 major coding benchmarks
GPQA: 86% - PhD-level scientific reasoning
AA-Intelligence Index: 52 - highest among Chinese LLMs
Significant SWE-bench Pro improvements over Qwen 3.6 Plus
Major advances in instruction following and world knowledge
Benchmark comparison
Qwen 3.6 Max Preview vs the Qwen 3.6 family
Qwen 3.6 Max Preview delivers flagship-class performance across coding, science, and general intelligence, with significant improvements over Qwen 3.6 Plus.
| Benchmark | Qwen 3.6 Max Preview Flagship MoE Featured | Qwen 3.6 Plus Proprietary | Qwen 3.6 27B Dense | Qwen 3.6 35B A3B MoE |
|---|---|---|---|---|
Coding benchmarks 6 major coding benchmarks | #1 | - | - | - |
GPQA PhD-level science | 86% | - | - | - |
AA-Intelligence Index General intelligence | 52 | - | - | - |
SWE-bench Pro Advanced software engineering Improvement over Plus | Improved | 56.6 | - | - |
SWE-bench Verified Real-world software engineering | - | 78.8% | 77.2% | 73.4% |
Terminal-Bench 2.0 Terminal operations | - | 61.6 | 59.3 | 51.5 |
Benchmark results from official Qwen 3.6 Max Preview release. Released April 20, 2026.
Coding Leader
Tops 6 major coding benchmarks - the definitive coding model
Qwen 3.6 Max Preview achieves the top position across 6 major coding benchmarks, with significant improvements over Qwen 3.6 Plus on SWE-bench Pro. From real-world GitHub issue resolution to competitive programming, it delivers the most capable coding assistance in the Qwen family.
- Tops 6 major coding benchmarks simultaneously
- Significant SWE-bench Pro improvements over Qwen 3.6 Plus
- Flagship-class code generation, debugging, and refactoring
Intelligence Leader
AA-Intelligence Index 52 - highest among Chinese LLMs
With an AA-Intelligence Index score of 52 and 86% on GPQA (PhD-level science), Qwen 3.6 Max Preview demonstrates the broadest and deepest general intelligence in the Qwen family. Major improvements in instruction following and world knowledge make it the most reliable model for complex, nuanced tasks.
- AA-Intelligence Index: 52 - highest among Chinese LLMs
- GPQA: 86% - PhD-level scientific reasoning
- Major advances in instruction following and world knowledge
Get started
Try Qwen 3.6 Max Preview now
Start chatting instantly or integrate via the OpenAI-compatible API.
Integration guides
Build with Qwen 3.6 Max Preview
Integrate the flagship model into your applications with OpenAI-compatible SDKs and agent frameworks.
Qwen ecosystem
The flagship of the Qwen 3.6 model family
Qwen 3.6 Max Preview is Alibaba's most capable model, leading the Qwen 3.6 family with top-tier coding, science, and general intelligence performance.
Get started
Ready to build with Qwen 3.6 Max Preview?
Start chatting instantly for free, or integrate via the OpenAI-compatible API with preserve_thinking for flagship-class agentic workflows.