DeepSeek-V4.1-Flash vs GPT-5.6 Luna
Wins 5 of 8 areas
Coding · Agents · Reasoning · Facts · Following instructions
Wins 2 of 8 areas
Images and charts · Math
DeepSeek-V4.1-Flash is the stronger all-rounder.GPT-5.6 Luna is cheaper and better at images and charts.
Scores updated · 37 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.
- AgentsCarrying out multi-step tasks on its own30DeepSeek-V4.1-Flash3 of 3 tests
- ReasoningHard problems that need careful thinking31DeepSeek-V4.1-Flash3 of 7 tests · 3 ties
- CodingWriting and fixing software32DeepSeek-V4.1-Flash3 of 5 tests
- FactsGetting facts right instead of making them up21DeepSeek-V4.1-Flash2 of 3 tests
- Following instructionsDoing exactly what it is asked10DeepSeek-V4.1-Flash1 of 1 test
- Images and chartsUnderstanding pictures, charts and video12GPT-5.6 Luna2 of 3 tests
- MathCompetition and research-level math12GPT-5.6 Luna2 of 3 tests
- Long documentsFinding answers in very long texts00Even0 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.
GPT-5.6 Luna costs 7% 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-V4.1-Flash pulls ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+28.9points ahead
- Common-sense trick questionsSimpleBench+19.9points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+18.5points ahead
Where GPT-5.6 Luna pulls ahead
- Reasoning about charts from research papersCharXiv (RQ)+13.7points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+12.2points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+10.9points ahead
Every test, side by side
All 37 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 28.977.348.4+28.9
- LMArena · WebDevDeepSeek-V4.1-Flash by 100 rating points16191519+100 rating
- Terminal-Bench 2.1DeepSeek-V4.1-Flash by 9.790.680.9+9.7
- LiveBench · CodingGPT-5.6 Luna by 2.98082.9+2.9
- SciCodeGPT-5.6 Luna by 1.751.953.6+1.7
AgentsDeepSeek-V4.1-Flash
- Terminal-Bench 4.0DeepSeek-V4.1-Flash by 15.226.811.6+15.2
- GDPValDeepSeek-V4.1-Flash by 75548+7
- AA IT-Bench SREDeepSeek-V4.1-Flash by 6.646.940.3+6.6
ReasoningDeepSeek-V4.1-Flash
- SimpleBenchDeepSeek-V4.1-Flash by 19.966.746.8+19.9
- ARC-AGI-2DeepSeek-V4.1-Flash by 18.578.159.6+18.5
- CritPtGPT-5.6 Luna by 6.314.320.6+6.3
- LiveBench · ReasoningDeepSeek-V4.1-Flash by 1.186.785.6+1.1
- Humanity's Last Examtie39.239.5tie
- GPQA Diamondtie90.991.1tie
- ARC-AGI-3tie0.20.2tie
FactsDeepSeek-V4.1-Flash
- AA-Omniscience · Non-hallucinationGPT-5.6 Luna by 3.93.57.4+3.9
- AA-Omniscience · AccuracyDeepSeek-V4.1-Flash by 3.746.442.7+3.7
- SimpleQA VerifiedDeepSeek-V4.1-Flash by 1.942.941+1.9
Images and chartsGPT-5.6 Luna
- CharXiv (RQ)GPT-5.6 Luna by 13.76982.7+13.7
- LMArena · VisionDeepSeek-V4.1-Flash by 21 rating points12781257+21 rating
- MMMU-ProGPT-5.6 Luna by 1.67778.6+1.6
MathGPT-5.6 Luna
- FrontierMath Tier 4GPT-5.6 Luna by 12.246.358.5+12.2
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Luna by 10.971.282.1+10.9
- LiveBench · MathematicsDeepSeek-V4.1-Flash by 6.193.387.2+6.1
Following instructionsDeepSeek-V4.1-Flash
- LiveBench · Instruction FollowingDeepSeek-V4.1-Flash by 9.97060.1+9.9
Other results11 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.
- Agents' Last ExamGPT-5.6 Luna by 18.531.850.3+18.5
- HLE (with tools)DeepSeek-V4.1-Flash by 1563.948.9+15
- Terminal-Bench 4.0DeepSeek-V4.1-Flash by 13.931.217.3+13.9
- OSWorld 2.0DeepSeek-V4.1-Flash by 8.754.345.6+8.7
- livebench_languageDeepSeek-V4.1-Flash by 8.681.272.6+8.6
- DeepSWE 1.1DeepSeek-V4.1-Flash by 7.274.267+7.2
- GDP (Surge AI)GPT-5.6 Luna by 5.517.222.7+5.5
- AA-OmniscienceDeepSeek-V4.1-Flash by 5-5.3-10.3+5
- ARC-AGI-1DeepSeek-V4.1-Flash by 2.893.590.7+2.8
- AA IntelligenceDeepSeek-V4.1-Flash by 2.239.537.3+2.2
- livebench_data_analysisDeepSeek-V4.1-Flash by 1.379.378+1.3
Questions people ask
Which is better, DeepSeek-V4.1-Flash or GPT-5.6 Luna?
DeepSeek-V4.1-Flash wins five of the eight areas we test: coding, agents, reasoning, facts and following instructions. GPT-5.6 Luna wins images and charts and math, and costs 7% less. They are level on long documents.
Which is better for coding?
DeepSeek-V4.1-Flash. It wins 3 of the 5 coding tests both models report; GPT-5.6 Luna wins 2.
Which is cheaper?
DeepSeek-V4.1-Flash costs $0.30 per million input tokens and $1.20 per million output tokens; GPT-5.6 Luna costs $0.20 and $1.20. That makes GPT-5.6 Luna about 7% cheaper for the same work.
How do you compare the two?
We use the 37 benchmark tests both models have published scores on. The verdict counts the 26 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 11 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.