Qwen3.6-27B release notes / what's new

Qwen has released Qwen3.6-27B, a dense 27-billion-parameter multimodal model that delivers flagship-level agentic coding performance. This model surpasses the previous-generation flagship Qwen3.5-397B-A17B across all major coding benchmarks, providing top-tier capabilities at a scale that is more practical for wide deployment than Mixture-of-Experts (MoE) architectures.

Agentic Coding Performance

Qwen3.6-27B achieves a breakthrough in agentic coding for dense models, outperforming models with up to 15x its total parameter count. It surpasses the Qwen3.5-397B-A17B (397B total / 17B active) on every major coding benchmark:

  • SWE-bench Verified: 77.2 vs. 76.2
  • SWE-bench Pro: 53.5 vs. 50.9
  • Terminal-Bench 2.0: 59.3 vs. 52.5
  • SkillsBench: 48.2 vs. 30.0

Additionally, the model is competitive with significantly larger models on reasoning tasks, scoring 87.8 on GPQA Diamond.

Multimodal Capabilities

Qwen3.6-27B is natively multimodal and supports both vision-language thinking and non-thinking modes within a single unified checkpoint. The model can process images and video alongside text to perform multimodal reasoning, visual question answering, and document understanding.

Key multimodal benchmarks include:

  • Video Understanding: 87.7 on VideoMME (with sub.) and 84.4 on VideoMMMU.
  • STEM & Puzzle: 82.9 on MMMU and 75.8 on MMMU-Pro.
  • Visual Agent: 94.7 on V*.
  • Document Understanding: 81.2 on CC-OCR and 78.4 on CharXiv RQ.

Deployment and API Integration

Qwen3.6-27B is available as open weights on Hugging Face and ModelScope, and accessible via Qwen Studio and the Alibaba Cloud Model Studio API.

API Features

The Model Studio API supports industry-standard protocols, including OpenAI-compatible chat completions and an interface compatible with Anthropic. A key feature of this release is preserve_thinking, which allows the model to preserve thinking content from all preceding turns in messages, a capability recommended for agentic tasks.

Third-Party Tool Compatibility

The model can be integrated with several AI coding assistants to enable context-aware development workflows:

  • OpenClaw: A self-hosted open-source AI coding agent.
  • Qwen Code: An open-source AI agent optimized for the Qwen series and designed for the terminal.
  • Claude Code: Supported via the Anthropic API protocol compatibility of Qwen APIs.

Summary of the Qwen3.6 Family

With the addition of Qwen3.6-27B, the Qwen3.6 open-source family provides a range of models for different scales, including the 3B-active Qwen3.6-35B-A3B, the API-accessible Qwen3.6-Plus, and the Qwen3.6-Max-Preview.

Sources