Across 34 shared benchmarks, DeepSeek-V3.2 scores higher on 16 and Qwen3.5 122B A10B on 18. The widest gap is τ³-Bench, where DeepSeek-V3.2 scores 69.2 against 13.6. DeepSeek-V3.2 is the cheaper of the two on tracked API pricing ($0.28 against $0.40 per million input tokens).
| Benchmark | DeepSeek-V3.2 | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 39.8 | 21.3 |
| AA Intelligence | 32.8 | 32.8 |
| AA-LCR | 70.7 | 70.3 |
| AA-Omniscience | -22.5 | -41.5 |
| Artificial Analysis Coding Index | 44.2 | 45.7 |
| browsecomp | 51.4 | 63.8 |
| browsecomp_zh | 65 | 69.9 |
| coding_arena_elo | 1368 | 1358 |
| critpt | 2.9 | 0.9 |
| gdpval | 18.8 | 24.3 |
| GPQA Diamond | 84 | 86.6 |
| HLE | 40.8 | 47.5 |
| HMMT 2025 | 90.2 | 90.3 |
| IFBench | 61 | 76.1 |
| LiveCodeBench v6 | 83.3 | 78.9 |
| longbench_v2 | 59.8 | 60.2 |
| mmlu_redux | 93.7 | 94 |
| MMLU-Pro | 86 | 86.7 |
| OmniScience Accuracy | 33 | 24.4 |
| OmniScience Non-Hallucination | 17.3 | 12.9 |
| scicode | 39 | 42 |
| Seal-0 | 49.5 | 44.1 |
| SWE-bench Verified | 73.1 | 72 |
| t2-bench | 80.3 | 79.5 |
| TauBench V3 - Banking | 18.8 | 15.3 |
| Terminal-Bench 2.0 | 46.4 | 49.4 |
| Terminal-Bench 2.1 | 46.8 | 47.6 |
| Terminal-Bench Hard | 35.6 | 31.1 |
| vectara_answer_rate | 92.6 | 99.8 |
| vectara_avg_summary_length | 62 | 86.4 |
| vectara_factual_consistency | 93.7 | 88.8 |
| vectara_hallucination_rate ↓ | 6.3 | 11.2 |
| τ²-Bench Telecom (AA run) | 90.6 | 93.6 |
| τ³-Bench | 69.2 | 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.