DeepSeek-V2.5 is the stronger all-rounder.
Scores updated · 21 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.
Reasoning and long documents 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-V2.5 pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+7points ahead
- Questions about very long textsLongBench v2+3.8points ahead
Where DeepSeek-V2 pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 21 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.
- HumanEvalDeepSeek-V2.5 by 45.743.389+45.7
- MATHDeepSeek-V2.5 by 31.343.474.7+31.3
- LiveCodeBenchDeepSeek-V2.5 by 30.211.641.8+30.2
- IFEvalDeepSeek-V2.5 by 22.957.780.6+22.9
- MATH-500 (EM)DeepSeek-V2.5 by 18.456.374.7+18.4
- CodeforcesDeepSeek-V2.5 by 18.117.535.6+18.1
- GSM8KDeepSeek-V2.5 by 15.979.295.1+15.9
- MMLU-ProDeepSeek-V2.5 by 14.851.466.2+14.8
- AIME 2024DeepSeek-V2.5 by 12.14.616.7+12.1
- Aider-Edit (Acc.)DeepSeek-V2.5 by 11.360.371.6+11.3
- CLUEWSCDeepSeek-V2.5 by 8.48290.4+8.4
- Chinese SimpleQA (C-SimpleQA)DeepSeek-V2.5 by 5.648.554.1+5.6
- BBHDeepSeek-V2.5 by 5.578.884.3+5.5
- DROP (3-shot F1)DeepSeek-V2.5 by 4.88387.8+4.8
- MMLUDeepSeek-V2.5 by 4.875.680.4+4.8
- HumanEval-Mul (Pass@1)DeepSeek-V2.5 by 4.569.373.8+4.5
- HellaSwagDeepSeek-V2.5 by 3.287.190.3+3.2
- MMLU-ReduxDeepSeek-V2.5 by 2.477.980.3+2.4
- FRAMES (Acc.)DeepSeek-V2 by 1.566.965.4+1.5
Questions people ask
Which is better, DeepSeek-V2 or DeepSeek-V2.5?
DeepSeek-V2.5 wins all two areas where both have results: reasoning and long documents. DeepSeek-V2 wins none.
How do you compare the two?
We use the 21 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 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.