DeepSeek-V3.2 vs GPT-4o
Wins 6 of 6 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
Wins 0 of 6 areas
—
DeepSeek-V3.2 is the stronger all-rounder.
Scores updated · 29 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.
- CodingWriting and fixing software40DeepSeek-V3.24 of 4 tests
- ReasoningHard problems that need careful thinking40DeepSeek-V3.24 of 4 tests
- FactsGetting facts right instead of making them up31DeepSeek-V3.23 of 4 tests
- Long documentsFinding answers in very long texts20DeepSeek-V3.22 of 2 tests
- AgentsCarrying out multi-step tasks on its own10DeepSeek-V3.21 of 1 test
- Following instructionsDoing exactly what it is asked10DeepSeek-V3.21 of 1 test
Agents 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-V3.2 costs 97% 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 DeepSeek-V3.2 pulls ahead
- Recent programming contest problemsLiveCodeBench v6+52.4points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+39.9points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+31.4points ahead
Where GPT-4o pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+44.8points ahead
Every test, side by side
All 29 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V3.2
- LiveCodeBench v6DeepSeek-V3.2 by 52.483.330.9+52.4
- SWE-bench VerifiedDeepSeek-V3.2 by 39.973.133.2+39.9
- Terminal-Bench HardDeepSeek-V3.2 by 27.335.68.3+27.3
- SciCodeDeepSeek-V3.2 by 5.638.933.3+5.6
ReasoningDeepSeek-V3.2
- GPQA DiamondDeepSeek-V3.2 by 31.48452.6+31.4
- Humanity's Last ExamDeepSeek-V3.2 by 22.824.61.8+22.8
- ARC-AGI-2DeepSeek-V3.2 by 440+4
- CritPtDeepSeek-V3.2 by 2.92.90+2.9
FactsDeepSeek-V3.2
- AA-Omniscience · Non-hallucinationGPT-4o by 44.817.362.1+44.8
- AA-Omniscience · AccuracyDeepSeek-V3.2 by 9.33323.7+9.3
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3.2 by 3.36.39.6+3.3
- SimpleQA VerifiedDeepSeek-V3.2 by 1.527.526+1.5
Long documentsDeepSeek-V3.2
- AA-LCRDeepSeek-V3.2 by 17.373.356+17.3
- LongBench v2DeepSeek-V3.2 by 11.759.848.1+11.7
Following instructionsDeepSeek-V3.2
- IFBenchDeepSeek-V3.2 by 24.760.736+24.7
Other results13 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 2025DeepSeek-V3.2 by 81.493.111.7+81.4
- τ²-Bench Telecom (AA run)DeepSeek-V3.2 by 61.790.628.9+61.7
- ARC-AGI-1DeepSeek-V3.2 by 52.5574.5+52.5
- LiveCodeBenchDeepSeek-V3.2 by 4583.338.3+45
- vectara_avg_summary_lengthGPT-4o by 24.66286.6+24.6
- Artificial Analysis Coding IndexDeepSeek-V3.2 by 2044.224.2+20
- AA IntelligenceDeepSeek-V3.2 by 14.221.57.3+14.2
- AA-OmniscienceGPT-4o by 12-22.5-10.5+12
- MMLU-ProDeepSeek-V3.2 by 10.38574.7+10.3
- AA Agentic IndexDeepSeek-V3.2 by 9.918.38.4+9.9
- MMLU-ReduxDeepSeek-V3.2 by 5.793.788+5.7
- vectara_factual_consistencyDeepSeek-V3.2 by 3.393.790.4+3.3
- vectara_answer_rateGPT-4o by 1.292.693.8+1.2
Questions people ask
Which is better, DeepSeek-V3.2 or GPT-4o?
DeepSeek-V3.2 wins all six areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. GPT-4o wins none.
Which is better for coding?
DeepSeek-V3.2. It wins 4 of the 4 coding tests both models report; GPT-4o wins none.
Which is cheaper?
DeepSeek-V3.2 costs $0.28 per million input tokens and $0.42 per million output tokens; GPT-4o costs $5.00 and $15.00. That makes DeepSeek-V3.2 about 97% cheaper for the same work.
How do you compare the two?
We use the 29 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 13 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.