| Benchmark | DeepSeek-R1 | Gemma 4 31B |
|---|---|---|
| AA Agentic Index | 3.1 | 14.4 |
| AA Intelligence | 18.6 | 29.7 |
| AA-LCR | 56 | 68.3 |
| AA-Omniscience | -31.3 | -47.9 |
| Artificial Analysis Coding Index | 24.6 | 43.4 |
| C-Eval | 91.8 | 82.6 |
| critpt | 0.6 | 1.4 |
| gdpval | 1.5 | 15.5 |
| GPQA Diamond | 81 | 85.7 |
| HLE | 17.7 | 26.5 |
| hmmt_feb_2025 | 76.7 | 88.7 |
| IFBench | 39 | 75.6 |
| mmlu_redux | 93.4 | 93.7 |
| MMLU-Pro | 85 | 85.2 |
| OmniScience Accuracy | 30.7 | 20 |
| OmniScience Non-Hallucination | 10.5 | 18.1 |
| scicode | 35.7 | 43.4 |
| simpleqa_verified | 27.4 | 9.6 |
| SWE-bench Multilingual | 30.5 | 51.7 |
| SWE-bench Verified | 57.6 | 52 |
| TauBench V3 - Banking | 6.4 | 14.8 |
| Terminal-Bench 2.1 | 19.1 | 43.4 |
| Terminal-Bench Hard | 6.1 | 36.4 |
| vectara_answer_rate | 97 | 100 |
| vectara_avg_summary_length | 93.5 | 75.8 |
| vectara_factual_consistency | 88.7 | 92.6 |
| vectara_hallucination_rate ↓ | 11.3 | 7.4 |
| τ²-Bench Telecom (AA run) | 11.4 | 65.5 |
| τ³-Bench | 6.4 | 67.5 |
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.