DeepSeek-V3.2-Exp vs GPT-5
Wins 1 of 6 areas
Facts
Wins 4 of 6 areas
Coding · Reasoning · Long documents · Following instructions
GPT-5 is the stronger all-rounder.DeepSeek-V3.2-Exp is cheaper and better at facts.
Scores updated · 32 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 software14GPT-54 of 5 tests
- ReasoningHard problems that need careful thinking03GPT-53 of 3 tests
- Long documentsFinding answers in very long texts01GPT-51 of 1 test
- Following instructionsDoing exactly what it is asked01GPT-51 of 1 test
- FactsGetting facts right instead of making them up21DeepSeek-V3.2-Exp2 of 3 tests
- AgentsCarrying out multi-step tasks on its own11Even1 each
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-V3.2-Exp costs 94% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where GPT-5 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+19points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+14.8points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+13.6points ahead
Where DeepSeek-V3.2-Exp pulls ahead
- Real work tasks from 44 professionsGDPVal+4points ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+2.6points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+1.4points ahead
Every test, side by side
All 32 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5
- LMArena · WebDevGPT-5 by 146 rating points12721418+146 rating
- SWE-bench VerifiedGPT-5 by 7.167.874.9+7.1
- SciCodeGPT-5 by 5.237.742.9+5.2
- SWE-bench MultilingualDeepSeek-V3.2-Exp by 2.657.955.3+2.6
- Terminal-Bench HardGPT-5 by 1.531.132.6+1.5
ReasoningGPT-5
- Humanity's Last ExamGPT-5 by 13.614.928.5+13.6
- GPQA DiamondGPT-5 by 5.779.785.4+5.7
- CritPtGPT-5 by 4.31.45.7+4.3
FactsDeepSeek-V3.2-Exp
- AA-Omniscience · AccuracyGPT-5 by 12.727.640.3+12.7
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3.2-Exp by 9.85.315.1+9.8
- AA-Omniscience · Non-hallucinationDeepSeek-V3.2-Exp by 1.419.217.8+1.4
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.
- τ²-Bench Telecom (AA run)GPT-5 by 50.933.984.8+50.9
- AA-OmniscienceGPT-5 by 22.3-31-8.7+22.3
- BrowseComp-ZHGPT-5 by 15.147.963+15.1
- Aider-PolyglotGPT-5 by 13.574.588+13.5
- vectara_avg_summary_lengthGPT-5 by 98.1 rating points64.6162.7+98.1 rating
- vectara_factual_consistencyDeepSeek-V3.2-Exp by 9.894.784.9+9.8
- HMMT 2025GPT-5 by 9.783.693.3+9.7
- LiveCodeBenchGPT-5 by 9.474.183.5+9.4
- AA IntelligenceGPT-5 by 6.416.623+6.4
- Terminal-BenchGPT-5 by 6.137.743.8+6.1
- AIME 2025GPT-5 by 5.389.394.6+5.3
- HMMT Nov. 2025GPT-5 by 584.289.2+5
- Artificial Analysis Coding IndexGPT-5 by 4.533.337.8+4.5
- vectara_answer_rateGPT-5 by 3.396.699.9+3.3
- AA Agentic IndexDeepSeek-V3.2-Exp by 2.228.726.5+2.2
- MMLU-ProGPT-5 by 2.18587.1+2.1
- HMMT Feb. 2025DeepSeek-V3.2-Exp by 1.79088.3+1.7
Questions people ask
Which is better, DeepSeek-V3.2-Exp or GPT-5?
GPT-5 wins four of the six areas where both have results: coding, reasoning, long documents and following instructions. DeepSeek-V3.2-Exp wins facts, and costs 94% less. They are level on agents.
Which is better for coding?
GPT-5. It wins 4 of the 5 coding tests both models report; DeepSeek-V3.2-Exp wins 1.
Which is cheaper?
DeepSeek-V3.2-Exp costs $0.28 per million input tokens and $0.42 per million output tokens; GPT-5 costs $1.25 and $10.00. That makes DeepSeek-V3.2-Exp about 94% cheaper for the same work.
How do you compare the two?
We use the 32 benchmark tests both models have published scores on. The verdict counts the 15 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.