Across 36 shared benchmarks, Gemini 3.1 Pro scores higher on 34 and Qwen3.5 122B A10B on 2. The widest gap is τ³-Bench, where Gemini 3.1 Pro scores 67.1 against 13.6. Qwen3.5 122B A10B is the cheaper of the two on tracked API pricing ($0.40 against $2.00 per million input tokens).
| Benchmark | Gemini 3.1 Pro | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 23 | 21.3 |
| AA Intelligence | 47.7 | 32.8 |
| AA-LCR | 79 | 70.3 |
| AA-Omniscience | 32.9 | -41.5 |
| arena_vision | 1294 | 1246 |
| Artificial Analysis Coding Index | 68.8 | 45.7 |
| baby_vision | 51.6 | 40.2 |
| browsecomp | 85.9 | 63.8 |
| coding_arena_elo | 1446 | 1358 |
| critpt | 17.7 | 0.9 |
| gdpval | 23.2 | 24.3 |
| GPQA Diamond | 94.3 | 86.6 |
| HLE | 51.4 | 47.5 |
| IFBench | 77.1 | 76.1 |
| LiveCodeBench v6 | 91.7 | 78.9 |
| mathvision | 89.8 | 86.2 |
| MMLU-Pro | 91 | 86.7 |
| mmmlu | 92.6 | 86.7 |
| MMMU-Pro | 83 | 76.9 |
| multichallenge | 71.4 | 61.5 |
| OmniScience Accuracy | 55.3 | 24.4 |
| OmniScience Non-Hallucination | 50.1 | 12.9 |
| OSWorld-Verified | 76.2 | 58 |
| scicode | 59 | 42 |
| SWE-bench Verified | 80.6 | 72 |
| t2-bench | 99.3 | 79.5 |
| TauBench V3 - Banking | 21.4 | 15.3 |
| Terminal-Bench 2.0 | 68.5 | 49.4 |
| Terminal-Bench 2.1 | 74 | 47.6 |
| Terminal-Bench Hard | 68.5 | 31.1 |
| vectara_answer_rate | 99.4 | 99.8 |
| vectara_avg_summary_length | 107.7 | 86.4 |
| vectara_factual_consistency | 89.6 | 88.8 |
| vectara_hallucination_rate ↓ | 10.4 | 11.2 |
| τ²-Bench Telecom (AA run) | 99.3 | 93.6 |
| τ³-Bench | 67.1 | 13.6 |
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.