DeepSeek-V4.1-Flash vs GPT-5.4
Wins 3 of 8 areas
Coding · Agents · Long documents
Wins 4 of 8 areas
Reasoning · Facts · Images and charts · Math
GPT-5.4 wins more areas, narrowly.DeepSeek-V4.1-Flash is cheaper and better at coding.
Scores updated · 40 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. 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 thinking14GPT-5.44 of 6 tests · 1 tie
- FactsGetting facts right instead of making them up03GPT-5.43 of 3 tests
- Images and chartsUnderstanding pictures, charts and video03GPT-5.43 of 3 tests
- MathCompetition and research-level math02GPT-5.42 of 3 tests · 1 tie
- CodingWriting and fixing software41DeepSeek-V4.1-Flash4 of 5 tests
- AgentsCarrying out multi-step tasks on its own20DeepSeek-V4.1-Flash2 of 2 tests
- Long documentsFinding answers in very long texts10DeepSeek-V4.1-Flash1 of 1 test
- Following instructionsDoing exactly what it is asked00Even0 each · 1 tie
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-V4.1-Flash costs 91% 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-5.4 pulls ahead
- Reasoning about charts from research papersCharXiv (RQ)+13.8points ahead
- Research-level physics problemsCritPt+9.1points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+7.4points ahead
Where DeepSeek-V4.1-Flash pulls ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+23.5points ahead
- Real work tasks from 44 professionsGDPVal+17.7points ahead
- Finds the root cause of IT system incidentsAA IT-Bench SRE+12.4points ahead
Every test, side by side
All 40 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V4.1-Flash
- LiveBench · Agentic CodingDeepSeek-V4.1-Flash by 23.577.353.8+23.5
- LMArena · WebDevDeepSeek-V4.1-Flash by 156 rating points16191463+156 rating
- Terminal-Bench 2.1DeepSeek-V4.1-Flash by 12.390.678.3+12.3
- SciCodeGPT-5.4 by 4.751.956.6+4.7
- LiveBench · CodingDeepSeek-V4.1-Flash by 2.58077.5+2.5
AgentsDeepSeek-V4.1-Flash
- GDPValDeepSeek-V4.1-Flash by 17.75537.3+17.7
- AA IT-Bench SREDeepSeek-V4.1-Flash by 12.446.934.5+12.4
ReasoningGPT-5.4
- CritPtGPT-5.4 by 9.114.323.4+9.1
- ARC-AGI-2DeepSeek-V4.1-Flash by 4.878.173.3+4.8
- Humanity's Last ExamGPT-5.4 by 4.539.243.7+4.5
- LiveBench · ReasoningGPT-5.4 by 1.486.788.1+1.4
- GPQA DiamondGPT-5.4 by 1.190.992+1.1
- ARC-AGI-3tie0.20.2tie
FactsGPT-5.4
- AA-Omniscience · Non-hallucinationGPT-5.4 by 4.83.58.3+4.8
- AA-Omniscience · AccuracyGPT-5.4 by 4.546.450.9+4.5
- SimpleQA VerifiedGPT-5.4 by 2.242.945.1+2.2
Images and chartsGPT-5.4
- CharXiv (RQ)GPT-5.4 by 13.86982.8+13.8
- LMArena · VisionGPT-5.4 by 25 rating points12781303+25 rating
- MMMU-ProGPT-5.4 by 1.47778.4+1.4
MathGPT-5.4
- FrontierMath Tiers 1-3 (v2)GPT-5.4 by 7.471.278.6+7.4
- FrontierMath Tier 4GPT-5.4 by 2.746.349+2.7
- LiveBench · Mathematicstie93.394.2tie
Following instructionsEven
- LiveBench · Instruction Followingtie7070.2tie
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.
- BabyVisionDeepSeek-V4.1-Flash by 39.989.649.7+39.9
- NL2RepoDeepSeek-V4.1-Flash by 22.76441.3+22.7
- DeepSWE 1.1DeepSeek-V4.1-Flash by 22.274.252+22.2
- CyberGymDeepSeek-V4.1-Flash by 21.888.166.3+21.8
- baby_vision_with_pythonDeepSeek-V4.1-Flash by 12.392.580.2+12.3
- HLE (with tools)DeepSeek-V4.1-Flash by 11.863.952.1+11.8
- AA-OmniscienceGPT-5.4 by 11.1-5.35.8+11.1
- charxiv_rq_with_pythonGPT-5.4 by 4.485.690+4.4
- vectara_factual_consistencyDeepSeek-V4.1-Flash by 2.895.893+2.8
- vectara_avg_summary_lengthDeepSeek-V4.1-Flash by 26.5 rating points108.281.7+26.5 rating
- livebench_languageGPT-5.4 by 1.481.282.6+1.4
- AA Intelligencetie39.539tie
- mmmu_pro_with_pythontie82.582.1tie
- ARC-AGI-1tie93.593.7tie
- livebench_data_analysistie79.379.3tie
- vectara_answer_ratetie99.999.9tie
Questions people ask
Which is better, DeepSeek-V4.1-Flash or GPT-5.4?
GPT-5.4 wins four of the eight areas we test: reasoning, facts, images and charts and math. DeepSeek-V4.1-Flash wins coding, agents and long documents, and costs 91% less. They are level on following instructions.
Which is better for coding?
DeepSeek-V4.1-Flash. It wins 4 of the 5 coding tests both models report; GPT-5.4 wins 1.
Which is cheaper?
DeepSeek-V4.1-Flash costs $0.30 per million input tokens and $1.20 per million output tokens; GPT-5.4 costs $2.50 and $15.00. That makes DeepSeek-V4.1-Flash about 91% cheaper for the same work.
How do you compare the two?
We use the 40 benchmark tests both models have published scores on. The verdict counts the 24 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.