DeepSeek-V3.1 vs GPT-5
Wins 0 of 6 areas
—
Wins 5 of 6 areas
Coding · Agents · Reasoning · Long documents · Following instructions
GPT-5 is the stronger all-rounder.DeepSeek-V3.1 is cheaper.
Scores updated · 33 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 thinking04GPT-54 of 4 tests
- CodingWriting and fixing software03GPT-53 of 4 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own02GPT-52 of 2 tests
- Following instructionsDoing exactly what it is asked02GPT-52 of 2 tests
- Long documentsFinding answers in very long texts01GPT-51 of 1 test
- FactsGetting facts right instead of making them up11Even1 each · 1 tie
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 costs 80% 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 GPT-5 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+31.6points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+24.9points ahead
- Keeps track of context across a multi-turn chatMulti-Challenge+23.5points ahead
Where DeepSeek-V3.1 pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 33 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5
- SWE-bench VerifiedGPT-5 by 8.96674.9+8.9
- Terminal-Bench HardGPT-5 by 7.62532.6+7.6
- SciCodeGPT-5 by 3.839.142.9+3.8
- SWE-bench Multilingualtie54.555.3tie
ReasoningGPT-5
- SimpleBenchGPT-5 by 16.74056.7+16.7
- Humanity's Last ExamGPT-5 by 14.214.328.5+14.2
- GPQA DiamondGPT-5 by 7.577.985.4+7.5
- CritPtGPT-5 by 3.725.7+3.7
FactsEven
- AA-Omniscience · AccuracyGPT-5 by 11.32940.3+11.3
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3.1 by 9.65.515.1+9.6
- AA-Omniscience · Non-hallucinationtie17.517.8tie
Following instructionsGPT-5
- IFBenchGPT-5 by 31.641.573.1+31.6
- Multi-ChallengeGPT-5 by 23.546.169.6+23.5
Other results17 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 2025GPT-5 by 59.833.593.3+59.8
- τ²-Bench Telecom (AA run)GPT-5 by 47.437.484.8+47.4
- AIME 2025GPT-5 by 44.849.894.6+44.8
- LiveCodeBenchGPT-5 by 27.156.483.5+27.1
- AA-OmniscienceGPT-5 by 20.9-29.6-8.7+20.9
- Aider-PolyglotGPT-5 by 19.668.488+19.6
- BrowseComp-ZHGPT-5 by 13.849.263+13.8
- Terminal-BenchGPT-5 by 12.531.343.8+12.5
- vectara_avg_summary_lengthGPT-5 by 99 rating points63.7162.7+99 rating
- vectara_factual_consistencyDeepSeek-V3.1 by 9.694.584.9+9.6
- AA IntelligenceGPT-5 by 9.513.523+9.5
- Artificial Analysis Coding IndexGPT-5 by 8.129.737.8+8.1
- AA Agentic IndexGPT-5 by 7.618.926.5+7.6
- vectara_answer_rateGPT-5 by 5.494.599.9+5.4
- MMLU-ReduxGPT-5 by 3.591.895.3+3.5
- MMLU-ProGPT-5 by 3.483.787.1+3.4
- HMMT Feb. 2025GPT-5 by 2.585.888.3+2.5
Questions people ask
Which is better, DeepSeek-V3.1 or GPT-5?
GPT-5 wins five of the six areas where both have results: coding, agents, reasoning, long documents and following instructions. DeepSeek-V3.1 wins none, but costs 80% less. They are level on facts.
Which is better for coding?
GPT-5. It wins 3 of the 4 coding tests both models report; DeepSeek-V3.1 wins none, and 1 is a tie.
Which is cheaper?
DeepSeek-V3.1 costs $0.56 per million input tokens and $1.68 per million output tokens; GPT-5 costs $1.25 and $10.00. That makes DeepSeek-V3.1 about 80% cheaper for the same work.
How do you compare the two?
We use the 33 benchmark tests both models have published scores on. The verdict counts the 16 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 17 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.