New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
A head-to-head benchmark comparison of Kimi K2 vs gemini-2.5-pro across 55 evaluations. gemini-2.5-pro leads on the most benchmarks (33).
Kimi K2 is our latest Mixture-of-Experts model with 32 billion activated parameters and 1 trillion total parameters. It achieves state-of-the-art performance in.
Full Kimi K2 specs →Gemini 2.5 Pro release date (per deprecations table)..
Full gemini-2.5-pro specs →| Benchmark | Kimi K2 | gemini-2.5-pro |
|---|---|---|
| agentic | ||
| BrowseComp | 14.1% | 7.6% |
| BrowseComp-Zh | 28.8 | — |
| FinSearchComp-T3 | 10.4 | — |
| GAIA | 57.7 | — |
| Seal-0 | 25.2 | — |
| TAU-bench Airline | 51.2% | 60% |
| TAU-bench Retail | 73.9% | 77% |
| Terminal-Bench 2.0 | 44.5% | 25.3% |
| WebWalkerQA | 63 | — |
| xbench-DeepSearch | 50 | — |
| code | ||
| Codeforces Elo | 1,574 | 1,837 |
| FullStackBench | 54% | 48.19% |
| HumanEval | 87.1% | — |
| LiveCodeBench | 48.95% | 80.1% |
| LiveCodeBench v5 | — | 69.1% |
| LiveCodeBench v6 | 56.1% | 70.65% |
| MBPP | 89.96% | 91.01% |
| SWE-bench Verified | 69.2% | 49% |
| SciCode | 30.7% | 42.8% |
| coding | ||
| Aider Benchmark | 85.34 | 94.36 |
| Multi-SWE-bench | 33.5 | — |
| MultiPL-E | 73.54 | 71.48 |
| OJ-Bench (cpp) | 25.5 | — |
| SWE-Bench Multilingual | 55.9 | — |
| general | ||
| ArenaHard v2 | 69.88 | 79.2 |
| C-Eval | 91.12% | 90.14% |
| IFEval | 90.99% | 90.8% |
| Longform Writing | 62.8 | — |
| MMLU | 89.5% | 91.9% |
| MMLU-Pro | 81.9% | 86.2% |
| MMLU-Redux | 92.7% | 94.67% |
| MultiChallenge | 54.1% | 57.5% |
| knowledge | ||
| AGIEval | — | 88.99 |
| CMMLU | — | 88.83 |
| TriviaQA | — | 84.45 |
| long-context | ||
| Frames | 72 | — |
| math | ||
| AIME 2024 | 67.24% | 92% |
| AIME 2025 | 75.2% | 88% |
| CNMO 2024 | 68.92% | 80.64% |
| HMMT February 2025 | 70.4% | 82.5% |
| IMOAnswerBench | 45.8 | — |
| MATH-500 | — | 96.7% |
| Omni-MATH | 62.42 | 82.14 |
| Omni-MATH-HARD | — | 69.36% |
| UGMathBench | 69.97 | 74.5 |
| reasoning | ||
| ARC-AGI-1 | 22.19 | 45.44 |
| BBEH | 34.83 | 51.51 |
| FinanceReasoning | 84.83 | 87.33 |
| GPQA Diamond | 74.2% | 86.4% |
| Humanity's Last Exam | 21.7% | 21.6% |
| ZebraLogic | 85.5% | 92.4% |
| safety | ||
| HealthBench | 43.8% | 49.39% |
| SimpleQA | 31% | 54% |
| science | ||
| PhyBench | — | 55.01 |
| tool-use | ||
| BFCL v3 | 71.1% | 63.31% |
Best result per row highlighted in cyan. Each benchmark links to its definition and sources; each model links to its full scorecard.
Across the 55 benchmarks compared here, gemini-2.5-pro leads on the most (33). The right choice still depends on your task — check the per-benchmark table above for coding, reasoning and math.
gemini-2.5-pro has the strongest result on Codeforces Elo among the models compared here. See the code-category rows in the table for the full picture.
This comparison covers 55 benchmarks on which at least one of the selected models has a published score.
Scores are aggregated from official model cards, technical reports and standard public evaluations, and link back to each benchmark's source. They are updated as new results are published.
Help us grow the database! Only the announcement link is required — we'll fetch the rest.
Missing a card? Drop the spec/announcement link — we'll verify the specs and add it. The link is the only required field.
Know an inference runtime we're missing (llama.cpp, vLLM, MLX, ComfyUI…)? Drop the link.
Spotted a wrong spec or an out-of-date price? Tell us and we'll verify it.
We'll send you a login code via email
