DeepSeek-V3 vs o3
Wins 0 of 6 areas
—
Wins 5 of 6 areas
Coding · Agents · Reasoning · Long documents · Following instructions
o3 is the stronger all-rounder.DeepSeek-V3 is cheaper.
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 thinking04o34 of 4 tests
- CodingWriting and fixing software03o33 of 3 tests
- Long documentsFinding answers in very long texts02o32 of 2 tests
- Following instructionsDoing exactly what it is asked02o32 of 2 tests
- AgentsCarrying out multi-step tasks on its own01o31 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
Agents 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 89% 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 o3 pulls ahead
- Common-sense trick questionsSimpleBench+34.2points ahead
- Reasons across sets of long documentsAA-LCR+34points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+30.4points ahead
Where DeepSeek-V3 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+2.2points 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.
Codingo3
- SWE-bench Verifiedo3 by 27.14269.1+27.1
- Terminal-Bench Hardo3 by 21.915.237.1+21.9
- SciCodeo3 by 23941+2
Reasoningo3
- SimpleBencho3 by 34.218.953.1+34.2
- GPQA Diamondo3 by 17.265.582.7+17.2
- Humanity's Last Examo3 by 15.44.720.1+15.4
- CritPto3 by 1.101.1+1.1
FactsEven
- AA-Omniscience · Accuracyo3 by 13.125.438.6+13.1
- AA-Omniscience · Non-hallucinationDeepSeek-V3 by 2.214.111.9+2.2
Long documentso3
- AA-LCRo3 by 3440.774.7+34
- LongBench v2o3 by 10.148.758.8+10.1
Following instructionso3
- IFBencho3 by 30.44171.4+30.4
- Multi-Challengeo3 by 2931.460.4+29
Other results24 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.
- AIME 2024o3 by 52.439.291.6+52.4
- livecodebench_hardo3 by 51.814.266+51.8
- HarmBencho3 by 48.749.798.4+48.7
- HMMT Feb. 2025o3 by 48.329.277.5+48.3
- AIR-Bench 2024o3 by 43.740.884.5+43.7
- livecodebench_mediumo3 by 36.353.589.8+36.3
- LiveCodeBencho3 by 35.549.284.7+35.5
- AIME 2025o3 by 35.151.386.4+35.1
- τ²-Bench Telecom (AA run)o3 by 33.647.180.7+33.6
- Aider-Polygloto3 by 31.749.681.3+31.7
- Tau2 airlineo3 by 25.83964.8+25.8
- AA-Omniscienceo3 by 25.2-40.7-15.6+25.2
- SimpleQAo3 by 24.524.949.4+24.5
- livecodebench_easyo3 by 15.883.399.1+15.8
- Artificial Analysis Coding Indexo3 by 15.42338.4+15.4
- ZebraLogico3 by 11.88495.8+11.8
- τ²-Bench (Retail)o3 by 11.169.180.2+11.1
- AA Intelligenceo3 by 10.59.720.2+10.5
- MMLU-Proo3 by 9.175.985+9.1
- MATH-500 (EM)o3 by 7.990.298.1+7.9
- simple_safety_testso3 by 3.795.399+3.7
- anthropic_red_teamo3 by 1.297.198.3+1.2
- bbqo3 by 1.296.797.9+1.2
- XSTesttie97.197.3tie
Questions people ask
Which is better, DeepSeek-V3 or o3?
o3 wins five of the six areas where both have results: coding, agents, reasoning, long documents and following instructions. DeepSeek-V3 wins none, but costs 89% less. They are level on facts.
Which is better for coding?
o3. It wins 3 of the 3 coding tests both models report; DeepSeek-V3 wins none.
Which is cheaper?
DeepSeek-V3 costs $0.24 per million input tokens and $0.90 per million output tokens; o3 costs $2.00 and $8.00. That makes DeepSeek-V3 about 89% 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 14 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 24 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.