Qwen3.6-27B
Qwen · released 2026-04-01 · Apache-2.0 license
Qwen3.6-27B is Alibaba's April 2026 open-weight dense multimodal model that delivers flagship-level coding (77.2% SWE-bench Verified) despite its small size. Hardest-reasoning breadth lags the largest models.
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | — |
| Architecture | — |
| Context window | — |
| Knowledge cutoff | — |
| Modalities | text, vision |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | 87.8% |
|---|---|
| SWE-bench Verified | 77.2% |
| AIME | 94.1% |
| MMLU-Pro | 86.2% |
| BFCL v3 (tool use) | — |
| Composite score | 7.37 |
| Community rating | No reviews yet |
Strengths & weaknesses
Strengths: Flagship-level coding from a 27B dense model (77.2 SWE-bench Verified); Strong GPQA (87.8) and LiveCodeBench (83.9) for its size; Multimodal, 262K native context, Apache-2.0
Weaknesses: HLE (24.0) below the largest frontier reasoners; Vendor scores use an internal agent scaffold