DeepSeek-V3 vs o1
Wins 1 of 6 areas
Coding
Wins 5 of 6 areas
Agents · Reasoning · Facts · Long documents · Following instructions
o1 is the stronger all-rounder.DeepSeek-V3 is cheaper and better at coding.
Scores updated · 38 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 thinking03o13 of 4 tests · 1 tie
- FactsGetting facts right instead of making them up02o12 of 2 tests
- AgentsCarrying out multi-step tasks on its own01o11 of 1 test
- Long documentsFinding answers in very long texts01o11 of 1 test
- Following instructionsDoing exactly what it is asked01o11 of 1 test
- CodingWriting and fixing software20DeepSeek-V32 of 3 tests · 1 tie
Agents, long documents and following instructions rest on a single test each.
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 98% 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 o1 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+29.3points ahead
- Reasons across sets of long documentsAA-LCR+24.3points ahead
- Common-sense trick questionsSimpleBench+21.2points ahead
Where DeepSeek-V3 pulls ahead
- Code for real scientific research problemsSciCode+3.2points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+2.3points ahead
Every test, side by side
All 38 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V3
- SciCodeDeepSeek-V3 by 3.23935.8+3.2
- Terminal-Bench HardDeepSeek-V3 by 2.315.212.9+2.3
- SWE-bench Verifiedtie4241.3tie
Reasoningo1
- SimpleBencho1 by 21.218.940.1+21.2
- GPQA Diamondo1 by 9.265.574.7+9.2
- Humanity's Last Examo1 by 2.34.77+2.3
- CritPttie00.3tie
Factso1
- AA-Omniscience · Non-hallucinationo1 by 16.314.130.4+16.3
- AA-Omniscience · Accuracyo1 by 925.434.5+9
Other results26 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 by 927 rating points11342061+927 rating
- Codeforces (Rating)o1 by 927 rating points11342061+927 rating
- HarmBencho1 by 46.649.796.3+46.6
- AIME 2024o1 by 4039.279.2+40
- AIR-Bench 2024o1 by 39.240.880+39.2
- Codeforces (Percentile)o1 by 37.958.796.6+37.9
- AIME 2025o1 by 30.451.381.7+30.4
- AA-Omniscienceo1 by 29.7-40.7-11.1+29.7
- HumanEvalo1 by 22.965.288.1+22.9
- LiveBenchDeepSeek-V3 by 20.172.452.3+20.1
- SimpleQAo1 by 17.524.942.4+17.5
- Artificial Analysis Coding Indexo1 by 16.72339.7+16.7
- τ²-Bench Telecom (AA run)o1 by 15.547.162.6+15.5
- Aider-Polygloto1 by 12.149.661.7+12.1
- MGSMo1 by 1179.890.8+11
- MMMLUo1 by 8.379.487.7+8.3
- MATHo1 by 6.290.296.4+6.2
- MATH-500 (EM)o1 by 6.290.296.4+6.2
- AA Intelligenceo1 by 5.59.715.2+5.5
- simple_safety_testso1 by 3.795.399+3.7
- MMLUo1 by 3.388.591.8+3.3
- GSM8Ko1 by 3.19497.1+3.1
- DROP (3-shot F1)DeepSeek-V3 by 1.491.690.2+1.4
- anthropic_red_teamo1 by 1.297.198.3+1.2
- bbqtie96.797.3tie
- XSTesttie97.197tie
Questions people ask
Which is better, DeepSeek-V3 or o1?
o1 wins five of the six areas where both have results: agents, reasoning, facts, long documents and following instructions. DeepSeek-V3 wins coding, and costs 98% less.
Which is better for coding?
DeepSeek-V3. It wins 2 of the 3 coding tests both models report; o1 wins none, and 1 is a tie.
Which is cheaper?
DeepSeek-V3 costs $0.24 per million input tokens and $0.90 per million output tokens; o1 costs $15.00 and $60.00. That makes DeepSeek-V3 about 98% cheaper for the same work.
How do you compare the two?
We use the 38 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 26 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.