DeepSeek-V4-Flash vs o3
Wins 6 of 7 areas
Coding · Agents · Reasoning · Math · Long documents · Following instructions
Wins 1 of 7 areas
Facts
DeepSeek-V4-Flash is the stronger all-rounder.o3 is better at facts.
Scores updated · 25 tests both models report · How we compare
Where each one wins
Tests won in each of the seven 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 thinking41DeepSeek-V4-Flash4 of 5 tests
- AgentsCarrying out multi-step tasks on its own20DeepSeek-V4-Flash2 of 2 tests
- CodingWriting and fixing software21DeepSeek-V4-Flash2 of 3 tests
- MathCompetition and research-level math10DeepSeek-V4-Flash1 of 1 test
- Long documentsFinding answers in very long texts10DeepSeek-V4-Flash1 of 1 test
- Following instructionsDoing exactly what it is asked10DeepSeek-V4-Flash1 of 1 test
- FactsGetting facts right instead of making them up12o32 of 3 tests
Math, 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-V4-Flash costs 82% 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-V4-Flash pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+54.9points ahead
- Real work tasks from 44 professionsGDPVal+34.1points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+24.2points ahead
Where o3 pulls ahead
- Short factual questions, answered correctlySimpleQA Verified+15.3points ahead
- Common-sense trick questionsSimpleBench+6.8points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+3.6points ahead
Every test, side by side
All 25 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V4-Flash
- SWE-bench VerifiedDeepSeek-V4-Flash by 9.97969.1+9.9
- SciCodeDeepSeek-V4-Flash by 9.350.341+9.3
- Terminal-Bench Hardo3 by 1.535.637.1+1.5
AgentsDeepSeek-V4-Flash
- GDPValDeepSeek-V4-Flash by 34.146.912.8+34.1
- BrowseCompDeepSeek-V4-Flash by 23.573.249.7+23.5
ReasoningDeepSeek-V4-Flash
- ARC-AGI-2DeepSeek-V4-Flash by 54.961.46.5+54.9
- Humanity's Last ExamDeepSeek-V4-Flash by 18.538.620.1+18.5
- CritPtDeepSeek-V4-Flash by 15.516.61.1+15.5
- GPQA DiamondDeepSeek-V4-Flash by 8.190.882.7+8.1
- SimpleBencho3 by 6.846.353.1+6.8
Factso3
- SimpleQA Verifiedo3 by 15.334.149.4+15.3
- AA-Omniscience · Non-hallucinationo3 by 3.68.311.9+3.6
- AA-Omniscience · AccuracyDeepSeek-V4-Flash by 1.840.438.6+1.8
MathDeepSeek-V4-Flash
- FrontierMath Tiers 1-3 (v2)DeepSeek-V4-Flash by 24.257.533.3+24.2
Following instructionsDeepSeek-V4-Flash
- IFBenchDeepSeek-V4-Flash by 7.879.271.4+7.8
Other results9 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.
- Artificial Analysis Coding IndexDeepSeek-V4-Flash by 30.769.138.4+30.7
- ARC-AGI-1DeepSeek-V4-Flash by 28.28960.8+28.2
- τ²-Bench Telecom (AA run)DeepSeek-V4-Flash by 14.39580.7+14.3
- AA IntelligenceDeepSeek-V4-Flash by 14.134.320.2+14.1
- LiveCodeBenchDeepSeek-V4-Flash by 6.991.684.7+6.9
- AA Agentic IndexDeepSeek-V4-Flash by 5.641.736.1+5.6
- AA-OmniscienceDeepSeek-V4-Flash by 1.2-14.3-15.6+1.2
- MMLU-ProDeepSeek-V4-Flash by 1.286.285+1.2
- StrongREJECTtie97.497tie
Questions people ask
Which is better, DeepSeek-V4-Flash or o3?
DeepSeek-V4-Flash wins six of the seven areas where both have results: coding, agents, reasoning, math, long documents and following instructions. o3 wins facts.
Which is better for coding?
DeepSeek-V4-Flash. It wins 2 of the 3 coding tests both models report; o3 wins 1.
Which is cheaper?
DeepSeek-V4-Flash costs $0.44 per million input tokens and $1.32 per million output tokens; o3 costs $2.00 and $8.00. That makes DeepSeek-V4-Flash about 82% cheaper for the same work.
How do you compare the two?
We use the 25 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 9 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.