Claude Sonnet 4 vs DeepSeek-V3.1-Terminus
Wins 1 of 6 areas
Long documents
Wins 2 of 6 areas
Agents · Reasoning
DeepSeek-V3.1-Terminus wins more areas, narrowly.Claude Sonnet 4 is better at long documents.
Scores updated · 20 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.
- ReasoningHard problems that need careful thinking03DeepSeek-V3.1-Terminus3 of 3 tests
- AgentsCarrying out multi-step tasks on its own01DeepSeek-V3.1-Terminus1 of 2 tests · 1 tie
- Long documentsFinding answers in very long texts10Claude Sonnet 41 of 1 test
- CodingWriting and fixing software11Even1 each · 1 tie
- FactsGetting facts right instead of making them up11Even1 each
- Following instructionsDoing exactly what it is asked11Even1 each
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-V3.1-Terminus costs 93% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where DeepSeek-V3.1-Terminus pulls ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+8.6points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+5.7points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+5points ahead
Where Claude Sonnet 4 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+45.7points ahead
- Keeps track of context across a multi-turn chatMulti-Challenge+2.7points ahead
- Code for real scientific research problemsSciCode+2points ahead
Every test, side by side
All 20 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingEven
- Terminal-Bench 2.1DeepSeek-V3.1-Terminus by 8.636.344.9+8.6
- SciCodeClaude Sonnet 4 by 24038+2
- Terminal-Bench Hardtie31.130.3tie
AgentsDeepSeek-V3.1-Terminus
- τ-Bench V3 · BankingDeepSeek-V3.1-Terminus by 4.316.721+4.3
- GDPValtie9.710.6tie
ReasoningDeepSeek-V3.1-Terminus
- Humanity's Last ExamDeepSeek-V3.1-Terminus by 5.710.716.4+5.7
- GPQA DiamondDeepSeek-V3.1-Terminus by 1.577.779.2+1.5
- CritPtDeepSeek-V3.1-Terminus by 1.40.31.7+1.4
FactsEven
- AA-Omniscience · Non-hallucinationClaude Sonnet 4 by 45.770.925.2+45.7
- AA-Omniscience · AccuracyDeepSeek-V3.1-Terminus by 522.727.7+5
Following instructionsEven
- Multi-ChallengeClaude Sonnet 4 by 2.757.154.4+2.7
- IFBenchDeepSeek-V3.1-Terminus by 2.354.757+2.3
Other results7 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.
- HMMT 2025DeepSeek-V3.1-Terminus by 70.215.986.1+70.2
- τ²-Bench Telecom (AA run)Claude Sonnet 4 by 27.564.637.1+27.5
- AA-OmniscienceClaude Sonnet 4 by 26.60.2-26.4+26.6
- AA Agentic IndexClaude Sonnet 4 by 8.717.68.9+8.7
- LiveCodeBenchDeepSeek-V3.1-Terminus by 6.468.574.9+6.4
- Artificial Analysis Coding IndexDeepSeek-V3.1-Terminus by 5.937.643.5+5.9
- AA IntelligenceClaude Sonnet 4 by 4.118.914.8+4.1
Questions people ask
Which is better, Claude Sonnet 4 or DeepSeek-V3.1-Terminus?
DeepSeek-V3.1-Terminus wins two of the six areas where both have results: agents and reasoning. Claude Sonnet 4 wins long documents. They are level on coding, facts and following instructions.
Which is better for coding?
Neither. They win 1 coding test each of the 3 both models report, and 1 is a tie.
Which is cheaper?
Claude Sonnet 4 costs $3.00 per million input tokens and $15.00 per million output tokens; DeepSeek-V3.1-Terminus costs $0.27 and $1.00. That makes DeepSeek-V3.1-Terminus about 93% cheaper for the same work.
How do you compare the two?
We use the 20 benchmark tests both models have published scores on. The verdict counts the 13 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 7 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.