Claude Opus 4 vs DeepSeek-V3
Wins 4 of 5 areas
Coding · Reasoning · Long documents · Following instructions
Wins 1 of 5 areas
Facts
Claude Opus 4 is the stronger all-rounder.DeepSeek-V3 is cheaper and better at facts.
Scores updated · 52 tests both models report · How we compare
Where each one wins
Tests won in each of the five 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.
- ReasoningHard problems that need careful thinking30Claude Opus 43 of 3 tests
- CodingWriting and fixing software20Claude Opus 42 of 4 tests · 2 ties
- Long documentsFinding answers in very long texts20Claude Opus 42 of 2 tests
- Following instructionsDoing exactly what it is asked20Claude Opus 42 of 2 tests
- FactsGetting facts right instead of making them up01DeepSeek-V31 of 1 test
Facts 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.
DeepSeek-V3 costs 99% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Claude Opus 4 pulls ahead
- Common-sense trick questionsSimpleBench+39.9points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+30.5points ahead
- Reasons across sets of long documentsAA-LCR+28.6points ahead
Where DeepSeek-V3 pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 52 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 4
- SWE-bench VerifiedClaude Opus 4 by 30.572.542+30.5
- Terminal-Bench HardClaude Opus 4 by 15.931.115.2+15.9
- SciCodetie39.839tie
- LiveCodeBench v6tie47.446.9tie
ReasoningClaude Opus 4
- SimpleBenchClaude Opus 4 by 39.958.818.9+39.9
- GPQA DiamondClaude Opus 4 by 14.179.665.5+14.1
- Humanity's Last ExamClaude Opus 4 by 7.612.34.7+7.6
FactsDeepSeek-V3
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3 by 5.9126.1+5.9
Long documentsClaude Opus 4
- AA-LCRClaude Opus 4 by 28.669.340.7+28.6
- LongBench v2Claude Opus 4 by 6.955.648.7+6.9
Following instructionsClaude Opus 4
- Multi-ChallengeClaude Opus 4 by 27.258.631.4+27.2
- IFBenchClaude Opus 4 by 12.753.741+12.7
Other results40 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.
- AIR-Bench 2024Claude Opus 4 by 44.985.740.8+44.9
- HarmBenchClaude Opus 4 by 4291.749.7+42
- HMMT Feb. 2025Claude Opus 4 by 30.86029.2+30.8
- livecodebench_mediumClaude Opus 4 by 26.980.453.5+26.9
- τ²-Bench Telecom (AA run)Claude Opus 4 by 26.373.447.1+26.3
- ZebraLogicDeepSeek-V3 by 24.759.384+24.7
- τ²-BenchClaude Opus 4 by 24.55732.5+24.5
- AIME 2025Claude Opus 4 by 24.275.551.3+24.2
- LiveCodeBenchClaude Opus 4 by 21.270.449.2+21.2
- Aider-PolyglotClaude Opus 4 by 21.170.749.6+21.1
- Tau2 airlineClaude Opus 4 by 216039+21
- livecodebench_hardClaude Opus 4 by 18.933.114.2+18.9
- livecodebench_easyClaude Opus 4 by 15.899.183.3+15.8
- CNMO 2024Claude Opus 4 by 14.457.643.2+14.4
- τ²-Bench (Retail)Claude Opus 4 by 12.781.869.1+12.7
- HMMT 2025DeepSeek-V3 by 11.615.927.5+11.6
- Artificial Analysis Coding IndexClaude Opus 4 by 113423+11
- AA IntelligenceClaude Opus 4 by 10.920.69.7+10.9
- MMLU-ProClaude Opus 4 by 10.786.675.9+10.7
- PolyMath-enDeepSeek-V3 by 9.749.859.5+9.7
- MMMLUClaude Opus 4 by 9.488.879.4+9.4
- AIME 2024Claude Opus 4 by 948.239.2+9
- MultiPL-EClaude Opus 4 by 6.589.683.1+6.5
- vectara_answer_rateDeepSeek-V3 by 6.59197.5+6.5
- vectara_factual_consistencyDeepSeek-V3 by 5.98893.9+5.9
- MMLU-ReduxClaude Opus 4 by 5.194.289.1+5.1
- simple_safety_testsClaude Opus 4 by 4.710095.3+4.7
- MMLUClaude Opus 4 by 4.492.988.5+4.4
- OJBenchDeepSeek-V3 by 4.419.624+4.4
- MATH-500 (EM)Claude Opus 4 by 4.294.490.2+4.2
- vectara_avg_summary_lengthClaude Opus 4 by 41.5 rating points123.281.7+41.5 rating
- AceBenchClaude Opus 4 by 2.975.672.7+2.9
- AutoLogiDeepSeek-V3 by 2.886.188.9+2.8
- SuperGPQAClaude Opus 4 by 2.856.553.7+2.8
- LiveBenchClaude Opus 4 by 2.274.672.4+2.2
- SimpleQADeepSeek-V3 by 2.122.824.9+2.1
- anthropic_red_teamClaude Opus 4 by 1.898.997.1+1.8
- IFEvalClaude Opus 4 by 1.387.486.1+1.3
- bbqtie97.196.7tie
- XSTesttie9797.1tie
Questions people ask
Which is better, Claude Opus 4 or DeepSeek-V3?
Claude Opus 4 wins four of the five areas where both have results: coding, reasoning, long documents and following instructions. DeepSeek-V3 wins facts, and costs 99% less.
Which is better for coding?
Claude Opus 4. It wins 2 of the 4 coding tests both models report; DeepSeek-V3 wins none, and 2 are ties.
Which is cheaper?
Claude Opus 4 costs $15.00 per million input tokens and $75.00 per million output tokens; DeepSeek-V3 costs $0.24 and $0.90. That makes DeepSeek-V3 about 99% cheaper for the same work.
How do you compare the two?
We use the 52 benchmark tests both models have published scores on. The verdict counts the 12 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 40 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.