The two are evenly matched.DeepSeek-V3.1-Base is better at reasoning; Kimi K2 Base at coding.
Scores updated · 41 tests both models report · How we compare
Where each one wins
Tests won in each of the two areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
Coding and reasoning rest on a single test each.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where DeepSeek-V3.1-Base pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+2.9points ahead
Where Kimi K2 Base pulls ahead
- Recent programming contest problemsLiveCodeBench v6+1.5points ahead
Every test, side by side
All 41 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Other results39 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- HumanEval+Kimi K2 Base by 20.264.684.8+20.2
- GSM-Infinite Hard (Tools-allowed)DeepSeek-V3.1-Base by 19.928.78.8+19.9
- GSM-Infinite Hard (128K)DeepSeek-V3.1-Base by 19.928.78.8+19.9
- GSM-Infinite Hard (64K)DeepSeek-V3.1-Base by 18.734.716+18.7
- MultiPL-E HumanEvalKimi K2 Base by 14.645.960.5+14.6
- MBPPKimi K2 Base by 14.474.689+14.4
- GSM-Infinite Hard (32K)DeepSeek-V3.1-Base by 12.738.826.1+12.7
- MMLU-ProKimi K2 Base by 10.458.869.2+10.4
- aime_2024_2025_combinedKimi K2 Base by 1021.631.6+10
- SimpleQAKimi K2 Base by 926.335.3+9
- MATHKimi K2 Base by 7.662.670.2+7.6
- GSM-Infinite Hard (16K)DeepSeek-V3.1-Base by 6.941.534.6+6.9
- CRUXEval-O (output prediction)DeepSeek-V3.1-Base by 6.876.469.6+6.8
- MultiPL-E MBPPKimi K2 Base by 6.352.558.8+6.3
- CRUXEval-I (input prediction)Kimi K2 Base by 5.962.168+5.9
- HumanEvalKimi K2 Base by 5.772.578.2+5.7
- HellaSwagKimi K2 Base by 5.489.294.6+5.4
- Chinese SimpleQA (C-SimpleQA)Kimi K2 Base by 3.770.974.6+3.7
- DROPDeepSeek-V3.1-Base by 2.786.383.6+2.7
- C-EvalKimi K2 Base by 2.59092.5+2.5
- SuperGPQAKimi K2 Base by 2.442.344.7+2.4
- NIAH-Multi (Tools-allowed)Kimi K2 Base by 2.397.299.5+2.3
- NIAH-Multi (128K)Kimi K2 Base by 2.397.299.5+2.3
- CMMLUKimi K2 Base by 2.188.890.9+2.1
- MBPP+Kimi K2 Base by 1.672.273.8+1.6
- MBPP+ (EvalPlus-augmented)Kimi K2 Base by 1.672.273.8+1.6
- TriviaQAKimi K2 Base by 1.683.585.1+1.6
- IncludeDeepSeek-V3.1-Base by 1.577.275.7+1.5
- NIAH-Multi (64K)Kimi K2 Base by 1.498.6100+1.4
- BigCodeBenchDeepSeek-V3.1-Base by 1.36361.7+1.3
- GlobalMMLUDeepSeek-V3.1-Base by 1.281.980.7+1.2
- GSM8Ktie91.492.1tie
- ARC-Challengetie95.696.2tie
- WinoGrandetie85.985.3tie
- BBHtie88.288.7tie
- MMLUtie87.487.8tie
- MMLU-Reduxtie9090.2tie
- NIAH-Multi (32K)tie99.799.8tie
- GPQA (unspecified)tie43.143.1tie
Questions people ask
Which is better, DeepSeek-V3.1-Base or Kimi K2 Base?
DeepSeek-V3.1-Base and Kimi K2 Base each win one of the two areas where both have results. DeepSeek-V3.1-Base wins reasoning; Kimi K2 Base wins coding.
Which is better for coding?
Kimi K2 Base. It wins the one coding test both models report.
How do you compare the two?
We use the 41 benchmark tests both models have published scores on. The verdict counts the 2 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 39 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.