DeepSeek-V3 is the stronger all-rounder.
Scores updated · 26 tests both models report · How we compare
Where each one wins
Tests won in each of the three 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, reasoning and long documents rest on a single test each.
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
- Fixes real GitHub issues in Python projectsSWE-bench Verified+25.2points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+23.2points ahead
- Questions about very long textsLongBench v2+13.3points ahead
Where DeepSeek-V2.5 pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 26 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Other results23 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.
- CodeforcesDeepSeek-V3 by 1098 rating points35.61134+1098 rating
- Aider-PolyglotDeepSeek-V3 by 31.817.849.6+31.8
- HumanEvalDeepSeek-V2.5 by 23.88965.2+23.8
- AIME 2024DeepSeek-V3 by 22.516.739.2+22.5
- MATHDeepSeek-V3 by 15.574.790.2+15.5
- MATH-500 (EM)DeepSeek-V3 by 15.574.790.2+15.5
- Arena HardDeepSeek-V3 by 15.276.291.4+15.2
- SimpleQADeepSeek-V3 by 14.710.224.9+14.7
- Chinese SimpleQA (C-SimpleQA)DeepSeek-V3 by 13.954.168+13.9
- MMLU-ProDeepSeek-V3 by 9.766.275.9+9.7
- HumanEval-Mul (Pass@1)DeepSeek-V3 by 8.873.882.6+8.8
- MMLU-ReduxDeepSeek-V3 by 8.880.389.1+8.8
- Aider-Edit (Acc.)DeepSeek-V3 by 8.171.679.7+8.1
- MMLUDeepSeek-V3 by 8.180.488.5+8.1
- FRAMES (Acc.)DeepSeek-V3 by 7.965.473.3+7.9
- LiveCodeBenchDeepSeek-V3 by 7.441.849.2+7.4
- IFEvalDeepSeek-V3 by 5.580.686.1+5.5
- DROP (3-shot F1)DeepSeek-V3 by 3.887.891.6+3.8
- BBHDeepSeek-V3 by 3.284.387.5+3.2
- AA IntelligenceDeepSeek-V3 by 3.16.69.7+3.1
- HellaSwagDeepSeek-V2.5 by 1.490.388.9+1.4
- GSM8KDeepSeek-V2.5 by 1.195.194+1.1
- CLUEWSCtie90.490.9tie
Questions people ask
Which is better, DeepSeek-V2.5 or DeepSeek-V3?
DeepSeek-V3 wins all three areas where both have results: coding, reasoning and long documents. DeepSeek-V2.5 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 26 benchmark tests both models have published scores on. The verdict counts the 3 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 23 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.