DeepSeek-V3 is the stronger all-rounder.
Scores updated · 18 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.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where DeepSeek-V3 pulls ahead
- Recent programming contest problemsLiveCodeBench v6+17.6points ahead
Where DeepSeek-V3-Base pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 18 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Other results17 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.
- MATHDeepSeek-V3 by 30.890.259.4+30.8
- MMLU-ProDeepSeek-V3 by 11.975.964+11.9
- SuperGPQADeepSeek-V3 by 7.553.746.2+7.5
- GSM8KDeepSeek-V3 by 5.89488.2+5.8
- DROPDeepSeek-V3 by 5.191.686.5+5.1
- C-EvalDeepSeek-V3-Base by 3.986.590.3+3.9
- MBPP+ (EvalPlus-augmented)DeepSeek-V3 by 3.378.875.5+3.3
- MMLU-ReduxDeepSeek-V3 by 2.389.186.8+2.3
- SimpleQADeepSeek-V3-Base by 1.324.926.1+1.3
- CRUXEval-O (output prediction)DeepSeek-V3-Base by 1.269.871+1.2
- CMMLUtie88.887.9tie
- MMLUtie88.587.7tie
- Chinese SimpleQA (C-SimpleQA)tie6868.7tie
- ARC-Challengetie95.394.7tie
- CRUXEval-I (input prediction)tie67.367.8tie
- PIQAtie84.784.2tie
- MMMLUtie79.479.5tie
Questions people ask
Which is better, DeepSeek-V3 or DeepSeek-V3-Base?
DeepSeek-V3 wins the one area where both have results: coding. DeepSeek-V3-Base wins none.
Which is better for coding?
DeepSeek-V3. It wins the one coding test both models report.
How do you compare the two?
We use the 18 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 17 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.