| Benchmark | DeepSeek-V4-Pro | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 63.3 | 54.6 |
| AA Intelligence | 53 | 34.6 |
| AA-LCR | 70 | 72.3 |
| AA-Omniscience | -10.6 | -44 |
| aime_2026 | 96.7 | 92.6 |
| Artificial Analysis Coding Index | 59.4 | 34.9 |
| browsecomp | 83.4 | 61 |
| coding_arena_elo | 1582 | 1357 |
| critpt | 13 | 0.9 |
| gdpval | 49 | 33 |
| GPQA Diamond | 90.5 | 85.8 |
| HLE | 48.2 | 48.5 |
| hmmt_feb_2026 | 95.2 | 84.3 |
| hmmt_nov_2025 | 94.4 | 89.8 |
| IFBench | 76.5 | 76.5 |
| imo_answer_bench | 89.8 | 79.9 |
| itbenchSre | 38.3 | 35.5 |
| MCP Atlas | 74.2 | 68.4 |
| MMLU-Pro | 87.5 | 86.1 |
| nl2repo | 38.5 | 27.3 |
| OmniScience Accuracy | 43 | 20.7 |
| OmniScience Non-Hallucination | 12.2 | 24.9 |
| scicode | 50 | 39.5 |
| SWE-bench Multilingual | 76.2 | 69.3 |
| SWE-bench Pro | 55.4 | 51.2 |
| SWE-bench Verified | 80.6 | 75 |
| Terminal-Bench 2.0 | 67.9 | 41.6 |
| Terminal-Bench Hard | 46.2 | 32.6 |
| tool_decathlon | 52.8 | 31.5 |
| vectara_answer_rate | 97.2 | 99.8 |
| vectara_avg_summary_length | 153.8 | 94.4 |
| vectara_factual_consistency | 91.4 | 87.9 |
| vectara_hallucination_rate ↓ | 8.6 | 12.1 |
| τ²-Bench Telecom (AA run) | 96.2 | 93.9 |
| τ³-Bench | 25.8 | 68.4 |
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.