Qwen 3.5 397B A17B
A strong general-purpose open model.
Benchmarks
| Benchmark | Qwen3.5-397B-A17B | gpt-oss-120b | Claude Opus 4.5 (closed model) |
|---|---|---|---|
| SWE-bench Verified | 76.4 | 62.4 | 80.9 |
| Terminal-Bench 2.0 | 52.5 | 18.7 | 59.3 |
| IFEval | 91.5 | 90.2 | 90.9 |
| AIME 2025 | 92.3 | 93.4 | 98.0 |
| LiveCodeBench v6 | 83.0 | 78.4 | 84.8 |
| MMLU-Pro | 87.6 | 79.7 | 89.5 |
| GPQA-Diamond | 88.4 | 78.9 | 87.0 |
| HLE | 27.5 | 18.3 | 30.8 |
Bold marks the higher score between the open models; Claude is included as a closed-model reference.
Sources: BenchGecko Qwen3.5 BenchGecko gpt-oss-120b BenchGecko Claude Opus 4.5 Qwen model card OpenAI model card Terminal-Bench BenchLM Claude Opus 4.5