DeepSeek-R1-Zero is the stronger all-rounder.
Scores updated · 16 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-R1-Zero pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+31points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+26.4points ahead
Where DeepSeek-V2.5 pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 16 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Other results14 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-R1-Zero by 1408 rating points35.61444+1408 rating
- AIME 2024DeepSeek-R1-Zero by 61.216.777.9+61.2
- IFEvalDeepSeek-V2.5 by 3480.646.6+34
- MATH-500 (EM)DeepSeek-R1-Zero by 21.274.795.9+21.2
- SimpleQADeepSeek-R1-Zero by 20.110.230.3+20.1
- FRAMES (Acc.)DeepSeek-R1-Zero by 16.965.482.3+16.9
- Chinese SimpleQA (C-SimpleQA)DeepSeek-R1-Zero by 12.354.166.4+12.3
- MMLUDeepSeek-R1-Zero by 8.480.488.8+8.4
- LiveCodeBenchDeepSeek-R1-Zero by 8.241.850+8.2
- Aider-PolyglotDeepSeek-V2.5 by 5.617.812.2+5.6
- MMLU-ReduxDeepSeek-R1-Zero by 5.380.385.6+5.3
- CLUEWSCDeepSeek-R1-Zero by 2.790.493.1+2.7
- MMLU-ProDeepSeek-R1-Zero by 2.766.268.9+2.7
- DROP (3-shot F1)DeepSeek-R1-Zero by 1.387.889.1+1.3
Questions people ask
Which is better, DeepSeek-V2.5 or DeepSeek-R1-Zero?
DeepSeek-R1-Zero wins all two areas where both have results: coding and reasoning. DeepSeek-V2.5 wins none.
Which is better for coding?
DeepSeek-R1-Zero. It wins the one coding test both models report.
How do you compare the two?
We use the 16 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 14 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.