Across 32 shared benchmarks, GPT-5.4 scores higher on 27 and Qwen3.5 122B A10B on 5. The widest gap is τ³-Bench, where GPT-5.4 scores 72.9 against 13.6. Qwen3.5 122B A10B is the cheaper of the two on tracked API pricing ($0.40 against $2.50 per million input tokens).
| Benchmark | GPT-5.4 | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 58.2 | 21.3 |
| AA Intelligence | 53.1 | 32.8 |
| AA-LCR | 77.7 | 70.3 |
| AA-Omniscience | 5.8 | -41.5 |
| arena_vision | 1293 | 1246 |
| Artificial Analysis Coding Index | 71.1 | 45.7 |
| baby_vision | 49.7 | 40.2 |
| browsecomp | 82.7 | 63.8 |
| coding_arena_elo | 1457 | 1358 |
| critpt | 23.4 | 0.9 |
| gdpval | 50.1 | 24.3 |
| GPQA Diamond | 93 | 86.6 |
| HLE | 43.7 | 47.5 |
| IFBench | 73.9 | 76.1 |
| mathvision | 92 | 86.2 |
| MMLU-Pro | 87.5 | 86.7 |
| MMMU-Pro | 81.2 | 76.9 |
| OmniDocBench 1.5 | 89.1 | 89.8 |
| OmniScience Accuracy | 50.9 | 24.4 |
| OmniScience Non-Hallucination | 17.4 | 12.9 |
| OSWorld-Verified | 75 | 58 |
| scicode | 56.6 | 42 |
| TauBench V3 - Banking | 39.6 | 15.3 |
| Terminal-Bench 2.0 | 75.1 | 49.4 |
| Terminal-Bench 2.1 | 78.3 | 47.6 |
| Terminal-Bench Hard | 57.6 | 31.1 |
| vectara_answer_rate | 99.9 | 99.8 |
| vectara_avg_summary_length | 81.7 | 86.4 |
| vectara_factual_consistency | 93 | 88.8 |
| vectara_hallucination_rate ↓ | 7 | 11.2 |
| τ²-Bench Telecom (AA run) | 87.1 | 93.6 |
| τ³-Bench | 72.9 | 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.