DeepSeek-V4-Flash vs GPT-5.2
Wins 3 of 7 areas
Agents · Reasoning · Following instructions
Wins 4 of 7 areas
Coding · Facts · Math · Long documents
GPT-5.2 wins more areas, narrowly.DeepSeek-V4-Flash is cheaper and better at reasoning.
Scores updated · 46 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.
- MathCompetition and research-level math15GPT-5.25 of 6 tests
- FactsGetting facts right instead of making them up03GPT-5.23 of 3 tests
- CodingWriting and fixing software35GPT-5.25 of 8 tests
- Long documentsFinding answers in very long texts01GPT-5.21 of 1 test
- ReasoningHard problems that need careful thinking30DeepSeek-V4-Flash3 of 6 tests · 3 ties
- AgentsCarrying out multi-step tasks on its own31DeepSeek-V4-Flash3 of 4 tests
- Following instructionsDoing exactly what it is asked20DeepSeek-V4-Flash2 of 2 tests
Long documents 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-V4-Flash costs 89% 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-5.2 pulls ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+11.4points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+10.5points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+9.9points ahead
Where DeepSeek-V4-Flash pulls ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+28.3points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+22.7points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+8.5points ahead
Every test, side by side
All 46 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.2
- SWE-bench ProDeepSeek-V4-Flash by 22.752.629.9+22.7
- Terminal-Bench HardGPT-5.2 by 11.435.647+11.4
- SWE-bench MultilingualDeepSeek-V4-Flash by 6.673.366.7+6.6
- LiveBench · Agentic CodingGPT-5.2 by 3.546.850.3+3.5
- SciCodeGPT-5.2 by 1.850.352.1+1.8
- LMArena · WebDevDeepSeek-V4-Flash by 15 rating points14301415+15 rating
- LiveBench · CodingGPT-5.2 by 1.17576.1+1.1
- SWE-bench VerifiedGPT-5.2 by 17980+1
AgentsDeepSeek-V4-Flash
- τ-Bench V3 · BankingDeepSeek-V4-Flash by 28.339.411.1+28.3
- MCP AtlasDeepSeek-V4-Flash by 8.46960.6+8.4
- BrowseCompDeepSeek-V4-Flash by 7.473.265.8+7.4
- GDPValGPT-5.2 by 1.446.948.3+1.4
ReasoningDeepSeek-V4-Flash
- ARC-AGI-2DeepSeek-V4-Flash by 8.561.452.9+8.5
- CritPtDeepSeek-V4-Flash by 516.611.6+5
- LiveBench · ReasoningDeepSeek-V4-Flash by 3.486.683.2+3.4
- Humanity's Last Examtie38.637.7tie
- GPQA Diamondtie90.890.3tie
- SimpleBenchtie46.345.8tie
FactsGPT-5.2
- AA-Omniscience · Non-hallucinationGPT-5.2 by 10.58.318.8+10.5
- AA-Omniscience · AccuracyGPT-5.2 by 3.940.444.3+3.9
- SimpleQA VerifiedGPT-5.2 by 334.137.1+3
MathGPT-5.2
- FrontierMath Tiers 1-3 (v2)GPT-5.2 by 9.957.567.4+9.9
- FrontierMath Tier 4GPT-5.2 by 7.324.431.7+7.3
- LiveBench · MathematicsGPT-5.2 by 6.486.893.2+6.4
- AIME 2026GPT-5.2 by 2.595.898.3+2.5
- HMMT Feb. 2026GPT-5.2 by 2.294.897+2.2
- IMOAnswerBenchDeepSeek-V4-Flash by 2.188.486.3+2.1
Following instructionsDeepSeek-V4-Flash
- IFBenchDeepSeek-V4-Flash by 3.879.275.4+3.8
- LiveBench · Instruction FollowingDeepSeek-V4-Flash by 3.765.561.8+3.7
Other results16 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.
- ToolathlonDeepSeek-V4-Flash by 2470.346.3+24
- Artificial Analysis Coding IndexDeepSeek-V4-Flash by 20.469.148.7+20.4
- AA Agentic IndexGPT-5.2 by 18.541.760.2+18.5
- AA-OmniscienceGPT-5.2 by 13.4-14.3-0.9+13.4
- τ²-Bench Telecom (AA run)DeepSeek-V4-Flash by 10.29584.8+10.2
- GDPval-AA (Elo)GPT-5.2 by 67 rating points13951462+67 rating
- AA IntelligenceDeepSeek-V4-Flash by 3.934.330.4+3.9
- GPQA (unspecified)GPT-5.2 by 3.988.592.4+3.9
- BrowseComp (context management)DeepSeek-V4-Flash by 3.273.270+3.2
- Terminal-Bench 2.0DeepSeek-V4-Flash by 2.956.954+2.9
- ARC-AGI-1DeepSeek-V4-Flash by 2.88986.2+2.8
- LiveCodeBenchDeepSeek-V4-Flash by 2.691.689+2.6
- livebench_data_analysisDeepSeek-V4-Flash by 1.179.378.2+1.1
- livebench_languagetie79.279.8tie
- MMLU-Protie86.286.7tie
- HLE (with tools)tie45.145.5tie
Questions people ask
Which is better, DeepSeek-V4-Flash or GPT-5.2?
GPT-5.2 wins four of the seven areas where both have results: coding, facts, math and long documents. DeepSeek-V4-Flash wins agents, reasoning and following instructions, and costs 89% less.
Which is better for coding?
GPT-5.2. It wins 5 of the 8 coding tests both models report; DeepSeek-V4-Flash wins 3.
Which is cheaper?
DeepSeek-V4-Flash costs $0.44 per million input tokens and $1.32 per million output tokens; GPT-5.2 costs $1.75 and $14.00. That makes DeepSeek-V4-Flash about 89% cheaper for the same work.
How do you compare the two?
We use the 46 benchmark tests both models have published scores on. The verdict counts the 30 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 16 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.