DeepSeek-V3 is the stronger all-rounder.
Scores updated · 36 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.
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
- Code for real scientific research problemsSciCode+6.7points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+5.2points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+1.1points ahead
Where o1-mini pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 36 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V3
- SciCodeDeepSeek-V3 by 6.73932.3+6.7
- SWE-bench Verifiedtie4241.6tie
- LiveCodeBench v6tie46.946.8tie
ReasoningDeepSeek-V3
- GPQA DiamondDeepSeek-V3 by 5.265.560.3+5.2
- Humanity's Last ExamDeepSeek-V3 by 1.14.73.6+1.1
- SimpleBenchtie18.918.1tie
Other results30 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.
- Codeforceso1-mini by 686 rating points11341820+686 rating
- Codeforces (Rating)o1-mini by 686 rating points11341820+686 rating
- HarmBencho1-mini by 38.849.788.5+38.8
- Codeforces (Percentile)o1-mini by 34.758.793.4+34.7
- Chinese SimpleQA (C-SimpleQA)DeepSeek-V3 by 27.76840.3+27.7
- HumanEvalo1-mini by 27.265.292.4+27.2
- AIME 2024o1-mini by 24.439.263.6+24.4
- CNMO 2024o1-mini by 24.443.267.6+24.4
- SimpleQADeepSeek-V3 by 17.924.97+17.9
- C-EvalDeepSeek-V3 by 17.686.568.9+17.6
- Aider-PolyglotDeepSeek-V3 by 16.749.632.9+16.7
- AlpacaEval2.0 (LC-winrate)DeepSeek-V3 by 12.27057.8+12.2
- SuperGPQADeepSeek-V3 by 8.553.745.2+8.5
- DROP (3-shot F1)DeepSeek-V3 by 7.791.683.9+7.7
- Arena-Hard (GPT-4-1106 judge)o1-mini by 6.585.592+6.5
- AIR-Bench 2024o1-mini by 4.540.845.3+4.5
- MMLU-Proo1-mini by 4.475.980.3+4.4
- FRAMES (Acc.)o1-mini by 3.673.376.9+3.6
- MMLUDeepSeek-V3 by 3.388.585.2+3.3
- MMLU-ReduxDeepSeek-V3 by 2.489.186.7+2.4
- simple_safety_testso1-mini by 1.795.397+1.7
- IFEvalDeepSeek-V3 by 1.386.184.8+1.3
- anthropic_red_teamo1-mini by 1.297.198.3+1.2
- CLUEWSCDeepSeek-V3 by 190.989.9+1
- AIME 2025tie51.350.7tie
- bbqtie96.796.9tie
- MATHtie90.290tie
- MATH-500 (EM)tie90.290tie
- AA Intelligencetie9.79.8tie
- XSTesttie97.197tie
Questions people ask
Which is better, DeepSeek-V3 or o1-mini?
DeepSeek-V3 wins all two areas where both have results: coding and reasoning. o1-mini wins none.
Which is better for coding?
DeepSeek-V3. It wins 1 of the 3 coding tests both models report; o1-mini wins none, and 2 are ties.
How do you compare the two?
We use the 36 benchmark tests both models have published scores on. The verdict counts the 6 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 30 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.