DeepSeek-V3-Base is the stronger all-rounder.
Scores updated · 20 tests both models report · How we compare
Where each one wins
Tests won in each of the one area 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 rests on a single test.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where DeepSeek-V3-Base pulls ahead
- Recent programming contest problemsLiveCodeBench v6+3points ahead
Where Kimi K2 Base pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 20 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Other results19 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.
- MATHKimi K2 Base by 10.859.470.2+10.8
- SimpleQAKimi K2 Base by 9.126.135.3+9.1
- Chinese SimpleQA (C-SimpleQA)Kimi K2 Base by 5.968.774.6+5.9
- CMathKimi K2 Base by 5.385.590.8+5.3
- MMLU-ProKimi K2 Base by 5.26469.2+5.2
- GSM8KKimi K2 Base by 3.988.292.1+3.9
- MMLU-ReduxKimi K2 Base by 3.486.890.2+3.4
- CMMLUKimi K2 Base by 387.990.9+3
- DROPDeepSeek-V3-Base by 2.986.583.6+2.9
- IncludeDeepSeek-V3-Base by 2.277.975.7+2.2
- C-EvalKimi K2 Base by 2.190.392.5+2.1
- MMMLUDeepSeek-V3-Base by 1.979.577.6+1.9
- MBPP+ (EvalPlus-augmented)DeepSeek-V3-Base by 1.775.573.8+1.7
- ARC-ChallengeKimi K2 Base by 1.594.796.2+1.5
- SuperGPQADeepSeek-V3-Base by 1.546.244.7+1.5
- CRUXEval-O (output prediction)DeepSeek-V3-Base by 1.47169.6+1.4
- PIQAtie84.284.9tie
- CRUXEval-I (input prediction)tie67.868tie
- MMLUtie87.787.8tie
Questions people ask
Which is better, DeepSeek-V3-Base or Kimi K2 Base?
DeepSeek-V3-Base wins the one area where both have results: coding. Kimi K2 Base wins none.
Which is better for coding?
DeepSeek-V3-Base. It wins the one coding test both models report.
How do you compare the two?
We use the 20 benchmark tests both models have published scores on. The verdict counts the 1 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 19 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.