| Benchmark | DeepSeek-V3 | Gemma 4 31B |
|---|---|---|
| AA Agentic Index | 1.6 | 14.4 |
| AA Intelligence | 15.4 | 29.7 |
| AA-LCR | 41.3 | 68.3 |
| AA-Omniscience | -37.6 | -47.9 |
| Artificial Analysis Coding Index | 23 | 43.4 |
| C-Eval | 90.1 | 82.6 |
| critpt | 0 | 1.4 |
| gdpval | 0 | 15.5 |
| GPQA Diamond | 68.4 | 85.7 |
| HLE | 5.2 | 26.5 |
| hmmt_feb_2025 | 29.2 | 88.7 |
| IFBench | 41 | 75.6 |
| LiveCodeBench v6 | 46.9 | 80 |
| mmlu_redux | 90.5 | 93.7 |
| MMLU-Pro | 81.2 | 85.2 |
| mmmlu | 79.4 | 88.4 |
| OmniScience Accuracy | 25.4 | 20 |
| OmniScience Non-Hallucination | 23.3 | 18.1 |
| scicode | 35.8 | 43.4 |
| supergpqa | 53.7 | 65.7 |
| SWE-bench Verified | 42 | 52 |
| TauBench V3 - Banking | 4.7 | 14.8 |
| Terminal-Bench 2.1 | 16.9 | 43.4 |
| Terminal-Bench Hard | 15.2 | 36.4 |
| vectara_answer_rate | 97.5 | 100 |
| vectara_avg_summary_length | 81.7 | 75.8 |
| vectara_factual_consistency | 93.9 | 92.6 |
| vectara_hallucination_rate ↓ | 6.1 | 7.4 |
| τ²-Bench | 32.5 | 76.9 |
| τ²-Bench Telecom (AA run) | 47.1 | 65.5 |
| τ³-Bench | 4.7 | 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.