| Benchmark | Gemini 2.5 Flash | Kimi K2 (Reasoning) |
|---|---|---|
| AA Agentic Index | 18.8 | 47.9 |
| AA Intelligence | 17.7 | 33.5 |
| AA-LCR | 65.7 | 70.3 |
| AA-Omniscience | -29.8 | -21.4 |
| aider_polyglot | 61.9 | 59.1 |
| AIME 2025 | 72 | 94.5 |
| Artificial Analysis Coding Index | 22.2 | 34.8 |
| critpt | 1.4 | 2.6 |
| frontiermath_tier_4 | 4.2 | 0 |
| gdpval | 11.9 | 24.5 |
| GPQA Diamond | 82.8 | 84.5 |
| HLE | 12.1 | 23.9 |
| hmmt_feb_2025 | 64.2 | 93.3 |
| IFBench | 50.3 | 68.1 |
| LiveCodeBench | 76.2 | 79.2 |
| mrcr | 32 | 44.2 |
| OmniScience Accuracy | 26.1 | 30.9 |
| OmniScience Non-Hallucination | 24.7 | 25.8 |
| scicode | 35.9 | 44.8 |
| simplebench | 41.2 | 39.6 |
| swe_bench_bash | 28.7 | 63.4 |
| SWE-bench Verified | 60.4 | 71.3 |
| Terminal-Bench Hard | 13.6 | 31.1 |
| vectara_answer_rate | 99 | 98.6 |
| vectara_avg_summary_length | 101.5 | 59.2 |
| vectara_factual_consistency | 92.2 | 82.1 |
| vectara_hallucination_rate ↓ | 7.8 | 17.9 |
| τ²-Bench Telecom (AA run) | 31.6 | 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.