Qwen3.6-35B-A3B
Open Weights
Alibaba Cloud
Alibaba's Apache-2.0 mixture-of-experts model, released April 2026, with 35B total parameters and only 3B activated per token. It accepts image, text and video input and emits text, natively handles a 262,144-token context (extensible to roughly 1.01M with YaRN scaling), and adds a thinking-preservation feature that carries reasoning context across conversation turns.
Strengths
- Only 3B of 35B parameters activate per token, giving strong efficiency for its capability
- Apache-2.0 licensed with no usage restrictions
- 262,144-token native context, extensible to roughly 1.01M tokens with YaRN
- Accepts image, text and video input
- 73.4 on SWE-bench Verified and 86.0 on GPQA
- Thinking preservation retains reasoning context across conversation turns
Caveats
- Text output only - the multimodality is input-side
- SWE-bench Pro (49.5) trails SWE-bench Verified (73.4) considerably on harder tasks
- Reaching the 1M context requires YaRN scaling rather than being native
- The model card states April 2026 but gives no exact release day
- Vendor
- Alibaba Cloud
- Context Window
- 262,144 tokens
- License
- Apache-2.0
- Modalities
- textimagevideo
| Benchmark | Score | Max |
|---|---|---|
| SWE-bench Verified | 73.4 | 100.0 |
| GPQA | 86.0 | 100.0 |
| VideoMME | 86.6 | 100.0 |
| MMLU-Pro | 85.2 | 100.0 |
| RealWorldQA | 85.3 | 100.0 |
| SWE-bench Pro | 49.5 | 100.0 |
Capabilities
Vision
Audio
Video
Tool Use
Resources
Pricing
Open weights under Apache-2.0 - self-host at no licence cost. Only 3B of 35B parameters activate per token, so inference is far cheaper than the total suggests.