Qwen3.6-35B-A3B
Multimodal
About
Qwen3.6-35B-A3B is an open-weight multimodal MoE model with 35B total parameters and 3B activated per token, released April 2026. It features a hybrid architecture combining Gated DeltaNet linear attention and standard Gated Attention with 256 total experts (8 routed + 1 shared), and includes a vision encoder for image and video understanding. Optimized for agentic coding, long-context reasoning, and visual tasks; supports 256K native context (extensible to ~1M via YaRN) with integrated thinking mode for multi-turn agent interactions.
Capabilities
VisionMultimodalReasoningFunction CallingTool UseStructured OutputsCode Execution
Benchmark Scores(6)
| Benchmark | Score | Version | Source |
|---|---|---|---|
| SWE-bench Verified | 73.4 | — | Qwen3.6-35B-A3B model card (April 2026) |
| SWE-bench Pro | 49.5 | — | Qwen3.6-35B-A3B model card (April 2026) |
| LiveCodeBench | 80.4 | v6 | Qwen3.6-35B-A3B model card (April 2026) |
| MMLU PRO | 85.2 | — | Qwen3.6-35B-A3B model card (April 2026) |
| Google-Proof Q&A | 86.0 | diamond | Qwen3.6-35B-A3B model card (April 2026) |
| MathVista | 86.4 | mini | Qwen3.6-35B-A3B model card (April 2026) |
