Across 21 shared benchmarks, Qwen2.5 Instruct 72B scores higher on 2 and Qwen3.5 122B A10B on 19. The widest gap is Terminal-Bench Hard, where Qwen3.5 122B A10B scores 31.1 against 4.5. Tracked API pricing per million tokens: Qwen2.5 Instruct 72B $0.47 in / $0.49 out, Qwen3.5 122B A10B $0.40 in / $3.20 out.
| Benchmark | Qwen2.5 Instruct 72B | Qwen3.5 122B A10B |
|---|---|---|
| AA Intelligence | 10 | 32.8 |
| AA-LCR | 20.3 | 70.3 |
| AA-Omniscience | -52.2 | -41.5 |
| Artificial Analysis Coding Index | 11.9 | 45.7 |
| C-Eval | 89.2 | 91.9 |
| critpt | 0 | 0.9 |
| GPQA Diamond | 49.1 | 86.6 |
| GSM8K | 95.8 | 94.5 |
| HLE | 4.2 | 47.5 |
| IFBench | 36.9 | 76.1 |
| ifeval | 87.2 | 93.4 |
| longbench_v2 | 39.4 | 60.2 |
| mmlu_redux | 86.8 | 94 |
| MMLU-Pro | 71.6 | 86.7 |
| mmmlu | 74.8 | 86.7 |
| OmniScience Accuracy | 17.6 | 24.4 |
| OmniScience Non-Hallucination | 15.3 | 12.9 |
| scicode | 26.7 | 42 |
| SWE-bench Verified | 23.8 | 72 |
| Terminal-Bench Hard | 4.5 | 31.1 |
| τ²-Bench Telecom (AA run) | 34.5 | 93.6 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing. Quoted rates are the price-setter row we currently track for each model — its direct or vendor-official listing where one exists (direct, direct), otherwise the lowest tracked offer.