Across 23 shared benchmarks, Gemini 2.5 Flash scores higher on 7 and Qwen3.5 122B A10B on 16. The widest gap is HLE, where Qwen3.5 122B A10B scores 47.5 against 12.1. Gemini 2.5 Flash is the cheaper of the two on tracked API pricing ($0.30 against $0.40 per million input tokens).
| Benchmark | Gemini 2.5 Flash | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 18.8 | 21.3 |
| AA Intelligence | 17.7 | 32.8 |
| AA-LCR | 65.7 | 70.3 |
| AA-Omniscience | -29.8 | -41.5 |
| arena_vision | 1236 | 1246 |
| Artificial Analysis Coding Index | 22.2 | 45.7 |
| critpt | 1.4 | 0.9 |
| gdpval | 11.9 | 24.3 |
| GPQA Diamond | 82.8 | 86.6 |
| HLE | 12.1 | 47.5 |
| IFBench | 50.3 | 76.1 |
| MMMU | 79.7 | 83.9 |
| MMMU-Pro | 69.1 | 76.9 |
| OmniScience Accuracy | 26.1 | 24.4 |
| OmniScience Non-Hallucination | 24.7 | 12.9 |
| scicode | 35.9 | 42 |
| SWE-bench Verified | 60.4 | 72 |
| Terminal-Bench Hard | 13.6 | 31.1 |
| vectara_answer_rate | 99 | 99.8 |
| vectara_avg_summary_length | 101.5 | 86.4 |
| vectara_factual_consistency | 92.2 | 88.8 |
| vectara_hallucination_rate ↓ | 7.8 | 11.2 |
| τ²-Bench Telecom (AA run) | 31.6 | 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.