Claude Sonnet 4 vs DeepSeek-V3
Wins 5 of 6 areas
Coding · Agents · Reasoning · Long documents · Following instructions
Wins 1 of 6 areas
Facts
Claude Sonnet 4 is the stronger all-rounder.DeepSeek-V3 is cheaper and better at facts.
Scores updated · 57 tests both models report · How we compare
Where each one wins
Tests won in each of the six 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.
- CodingWriting and fixing software50Claude Sonnet 45 of 5 tests
- ReasoningHard problems that need careful thinking30Claude Sonnet 43 of 4 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own20Claude Sonnet 42 of 2 tests
- Following instructionsDoing exactly what it is asked20Claude Sonnet 42 of 2 tests
- Long documentsFinding answers in very long texts10Claude Sonnet 41 of 1 test
- FactsGetting facts right instead of making them up12DeepSeek-V32 of 3 tests
Long documents 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 94% 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 Sonnet 4 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+56.8points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+30.7points ahead
- Reasons across sets of long documentsAA-LCR+29.6points ahead
Where DeepSeek-V3 pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+2.8points ahead
Every test, side by side
All 57 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4
- SWE-bench VerifiedClaude Sonnet 4 by 30.772.742+30.7
- Terminal-Bench 2.1Claude Sonnet 4 by 19.436.316.9+19.4
- Terminal-Bench HardClaude Sonnet 4 by 15.931.115.2+15.9
- LiveCodeBench v6Claude Sonnet 4 by 1.648.546.9+1.6
- SciCodeClaude Sonnet 4 by 14039+1
AgentsClaude Sonnet 4
- τ-Bench V3 · BankingClaude Sonnet 4 by 1216.74.7+12
- GDPValClaude Sonnet 4 by 9.79.70+9.7
ReasoningClaude Sonnet 4
- SimpleBenchClaude Sonnet 4 by 26.645.518.9+26.6
- GPQA DiamondClaude Sonnet 4 by 12.277.765.5+12.2
- Humanity's Last ExamClaude Sonnet 4 by 610.74.7+6
- CritPttie0.30tie
FactsDeepSeek-V3
- AA-Omniscience · Non-hallucinationClaude Sonnet 4 by 56.870.914.1+56.8
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3 by 4.210.36.1+4.2
- AA-Omniscience · AccuracyDeepSeek-V3 by 2.822.725.4+2.8
Following instructionsClaude Sonnet 4
- Multi-ChallengeClaude Sonnet 4 by 25.757.131.4+25.7
- IFBenchClaude Sonnet 4 by 13.754.741+13.7
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.
- HarmBenchClaude Sonnet 4 by 48.39849.7+48.3
- AIR-Bench 2024Claude Sonnet 4 by 47.588.340.8+47.5
- AA-OmniscienceClaude Sonnet 4 by 40.90.2-40.7+40.9
- livecodebench_mediumClaude Sonnet 4 by 22.57653.5+22.5
- Aider-PolyglotClaude Sonnet 4 by 21.170.749.6+21.1
- LiveCodeBenchClaude Sonnet 4 by 19.368.549.2+19.3
- AIME 2025Claude Sonnet 4 by 19.270.551.3+19.2
- livecodebench_hardClaude Sonnet 4 by 18.432.614.2+18.4
- τ²-Bench Telecom (AA run)Claude Sonnet 4 by 17.564.647.1+17.5
- CNMO 2024Claude Sonnet 4 by 17.260.443.2+17.2
- Tau2 airlineClaude Sonnet 4 by 16.555.539+16.5
- livecodebench_easyClaude Sonnet 4 by 15.598.883.3+15.5
- Artificial Analysis Coding IndexClaude Sonnet 4 by 14.637.623+14.6
- τ²-BenchClaude Sonnet 4 by 12.745.232.5+12.7
- HMMT 2025DeepSeek-V3 by 11.615.927.5+11.6
- ZebraLogicDeepSeek-V3 by 10.373.784+10.3
- AA IntelligenceClaude Sonnet 4 by 9.218.99.7+9.2
- SimpleQADeepSeek-V3 by 915.924.9+9
- OJBenchDeepSeek-V3 by 8.715.324+8.7
- MMLU-ProClaude Sonnet 4 by 7.883.775.9+7.8
- MMMLUClaude Sonnet 4 by 7.186.579.4+7.1
- PolyMath-enDeepSeek-V3 by 6.752.859.5+6.7
- vectara_avg_summary_lengthClaude Sonnet 4 by 64.1 rating points145.881.7+64.1 rating
- τ²-Bench (Retail)Claude Sonnet 4 by 5.97569.1+5.9
- MultiPL-EClaude Sonnet 4 by 5.588.683.1+5.5
- simple_safety_testsClaude Sonnet 4 by 4.710095.3+4.7
- MMLU-ReduxClaude Sonnet 4 by 4.593.689.1+4.5
- AIME 2024Claude Sonnet 4 by 4.243.439.2+4.2
- vectara_factual_consistencyDeepSeek-V3 by 4.289.793.9+4.2
- MATH-500 (EM)Claude Sonnet 4 by 3.89490.2+3.8
- AceBenchClaude Sonnet 4 by 3.576.272.7+3.5
- MMLUClaude Sonnet 4 by 391.588.5+3
- LiveBenchClaude Sonnet 4 by 2.474.872.4+2.4
- anthropic_red_teamClaude Sonnet 4 by 2.199.297.1+2.1
- SuperGPQAClaude Sonnet 4 by 255.753.7+2
- IFEvalClaude Sonnet 4 by 1.587.686.1+1.5
- vectara_answer_rateClaude Sonnet 4 by 1.198.697.5+1.1
- AutoLogitie89.888.9tie
- XSTesttie96.697.1tie
- bbqtie96.696.7tie
Questions people ask
Which is better, Claude Sonnet 4 or DeepSeek-V3?
Claude Sonnet 4 wins five of the six areas where both have results: coding, agents, reasoning, long documents and following instructions. DeepSeek-V3 wins facts, and costs 94% less.
Which is better for coding?
Claude Sonnet 4. It wins 5 of the 5 coding tests both models report; DeepSeek-V3 wins none.
Which is cheaper?
Claude Sonnet 4 costs $3.00 per million input tokens and $15.00 per million output tokens; DeepSeek-V3 costs $0.24 and $0.90. That makes DeepSeek-V3 about 94% cheaper for the same work.
How do you compare the two?
We use the 57 benchmark tests both models have published scores on. The verdict counts the 17 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.