Moonshot AI·released 2025-11-061 source
Kimi K2 Thinking benchmark scores: 9 benchmarks tracked. 56% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 9 benchmarks·best result 83.04 on OTIS Mock AIME 2024-2025 (reproduced)·5 of 9 independently reproduced·$0.6/$2.5 per M tokens
Consensus: LiteLLM · Cross-check: models.dev, OpenRouter · See every model’s pricing →
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 83.04accuracy (%) | reproduced· optimizedT1 | 2025-11-06Epoch AI |
| GPQA diamond | 78.96accuracy (%) | reproduced· optimizedT1 | 2025-11-06Epoch AI |
| METR Time Horizons | 59.19accuracy (%) | unverifiedT1 | 2025-11-06METR - Measuring AI Ability to Complete Long Tasks |
| WeirdML | 42.79accuracy (%) | unverifiedT1 | 2025-11-06https://htihle.github.io/weirdml.html |
| FrontierMath-2025-02-28-Private | 37.55accuracy (%) | reproduced· optimizedT1 | 2025-11-06Epoch AI |
| Terminal Bench | 35.73accuracy (%) | unverifiedT1 | 2025-11-06https://www.tbench.ai/leaderboard/terminal-bench/2.0 |
| CL-bench | 17.6accuracy (%) | unverifiedT2 | 2025-11-06 |
| Chess Puzzles | 15.82accuracy (%) | reproducedT1 | 2025-11-06Epoch AI |
| FrontierMath-Tier-4-2025-07-01-Private | 0accuracy (%) | reproducedT1 | 2025-11-06Epoch AI |
Claims drawn from cited facts, not live model generation.
This model originates from Moonshot AI and was released in 2025, making it a recent arrival in the current generation of models. Because no parameter count accompanies it, its scale is not pinned down here, so its positioning rests on that recency rather than on a stated size.
2 cited facts
This model is scored on nine tracked benchmarks. It holds no current top score on any of them. Its record can be trusted as outside evaluation has independently reproduced the scores.
3 cited facts
Its mean gap to the best recorded score across its tracked benchmarks is 31.32 points.
1 cited fact
6 facts cross-checked across data sources: 3 corroborated, 1 single-source, 2 disputed.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks, litellm_prices, modelsdev_models, openrouter_models
Kimi K2 Thinking is an AI model developed by Moonshot AI, released 2025-11-06. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
Kimi K2 Thinking has recorded scores on 9 benchmarks, each shown with its evidence status.
Kimi K2 Thinking has recorded scores on 9 benchmarks — GPQA diamond, OTIS Mock AIME 2024-2025, WeirdML, Chess Puzzles, METR Time Horizons, FrontierMath-2025-02-28-Private, and 3 more. The full table above shows each score with its evidence status.
5 of 9 recorded scores (56%) are independently reproduced rather than self-reported by the lab.
Kimi K2 Thinking has 9 tracked claims: 5 independently reproduced, 4 unverified.
Listed API pricing: $0.6 per million input tokens, $2.5 per million output tokens. See the pricing block for the full breakdown.