| Benchmark | Gemma 4 31B | Kimi K2 (Reasoning) |
|---|---|---|
| AA Agentic Index | 14.4 | 47.9 |
| AA Intelligence | 29.7 | 33.5 |
| AA-LCR | 68.3 | 70.3 |
| AA-Omniscience | -47.9 | -21.4 |
| Artificial Analysis Coding Index | 43.4 | 34.8 |
| critpt | 1.4 | 2.6 |
| gdpval | 15.5 | 24.5 |
| GPQA (unspecified) | 84.3 | 74.2 |
| GPQA Diamond | 85.7 | 84.5 |
| HLE | 26.5 | 23.9 |
| HLE (with tools) | 26.5 | 44.9 |
| hmmt_feb_2025 | 88.7 | 93.3 |
| hmmt_nov_2025 | 87.5 | 89.2 |
| IFBench | 75.6 | 68.1 |
| imo_answer_bench | 74.5 | 78.6 |
| LiveCodeBench v6 | 80 | 83.1 |
| mmlu_redux | 93.7 | 94.4 |
| MMLU-Pro | 85.2 | 84.6 |
| OmniScience Accuracy | 20 | 30.9 |
| OmniScience Non-Hallucination | 18.1 | 25.8 |
| scicode | 43.4 | 44.8 |
| SWE-bench Multilingual | 51.7 | 61.1 |
| SWE-bench Verified | 52 | 71.3 |
| Terminal-Bench 2.0 | 42.9 | 35.7 |
| Terminal-Bench Hard | 36.4 | 31.1 |
| vectara_answer_rate | 100 | 98.6 |
| vectara_avg_summary_length | 75.8 | 59.2 |
| vectara_factual_consistency | 92.6 | 82.1 |
| vectara_hallucination_rate ↓ | 7.4 | 17.9 |
| τ²-Bench | 76.9 | 74.3 |
| τ²-Bench Telecom (AA run) | 65.5 | 93 |
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.