DeepSeek-V4-Flash vs GPT-5.6 Luna
Wins 2 of 7 areas
Reasoning · Following instructions
Wins 5 of 7 areas
Coding · Agents · Facts · Math · Long documents
GPT-5.6 Luna is the stronger all-rounder.DeepSeek-V4-Flash is cheaper and better at reasoning.
Scores updated · 45 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.
- CodingWriting and fixing software07GPT-5.6 Luna7 of 7 tests
- MathCompetition and research-level math04GPT-5.6 Luna4 of 5 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own13GPT-5.6 Luna3 of 5 tests · 1 tie
- FactsGetting facts right instead of making them up02GPT-5.6 Luna2 of 3 tests · 1 tie
- Long documentsFinding answers in very long texts01GPT-5.6 Luna1 of 1 test
- ReasoningHard problems that need careful thinking21DeepSeek-V4-Flash2 of 6 tests · 3 ties
- Following instructionsDoing exactly what it is asked10DeepSeek-V4-Flash1 of 1 test
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-Flash costs 71% 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-5.6 Luna pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+34.1points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+24.6points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+14points ahead
Where DeepSeek-V4-Flash pulls ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+8.3points ahead
- Follows detailed instructions exactlyLiveBench · Instruction Following+5.4points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+1.8points ahead
Every test, side by side
All 45 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.6 Luna
- SWE-bench VerifiedGPT-5.6 Luna by 147993+14
- SWE-bench ProGPT-5.6 Luna by 10.152.662.7+10.1
- LMArena · WebDevGPT-5.6 Luna by 88 rating points14311519+88 rating
- LiveBench · CodingGPT-5.6 Luna by 7.97582.9+7.9
- SciCodeGPT-5.6 Luna by 3.350.353.6+3.3
- Terminal-Bench 2.1GPT-5.6 Luna by 2.278.780.9+2.2
- LiveBench · Agentic CodingGPT-5.6 Luna by 1.646.848.4+1.6
AgentsGPT-5.6 Luna
- BrowseCompGPT-5.6 Luna by 10.173.283.3+10.1
- AA IT-Bench SREGPT-5.6 Luna by 8.831.540.3+8.8
- τ-Bench V3 · BankingDeepSeek-V4-Flash by 8.339.431.1+8.3
- GDPValGPT-5.6 Luna by 1.346.748+1.3
- Terminal-Bench 4.0tie12.111.6tie
ReasoningDeepSeek-V4-Flash
- CritPtGPT-5.6 Luna by 416.620.6+4
- ARC-AGI-2DeepSeek-V4-Flash by 1.861.459.6+1.8
- LiveBench · ReasoningDeepSeek-V4-Flash by 186.685.6+1
- Humanity's Last Examtie38.639.5tie
- SimpleBenchtie46.346.8tie
- GPQA Diamondtie90.891.1tie
FactsGPT-5.6 Luna
- SimpleQA VerifiedGPT-5.6 Luna by 6.934.141+6.9
- AA-Omniscience · AccuracyGPT-5.6 Luna by 2.340.442.7+2.3
- AA-Omniscience · Non-hallucinationtie8.37.4tie
MathGPT-5.6 Luna
- FrontierMath Tier 4GPT-5.6 Luna by 34.124.458.5+34.1
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Luna by 24.657.582.1+24.6
- HMMT Feb. 2026GPT-5.6 Luna by 3.794.898.5+3.7
- AIME 2026GPT-5.6 Luna by 1.895.897.6+1.8
- LiveBench · Mathematicstie86.887.2tie
Following instructionsDeepSeek-V4-Flash
- LiveBench · Instruction FollowingDeepSeek-V4-Flash by 5.465.560.1+5.4
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.
- GDPval-AA v2GPT-5.6 Luna by 341 rating points11891530+341 rating
- Agents' Last ExamGPT-5.6 Luna by 25.125.250.3+25.1
- ToolathlonDeepSeek-V4-Flash by 16.970.353.4+16.9
- DeepSWEGPT-5.6 Luna by 12.854.467.2+12.8
- DeepSWE 1.1GPT-5.6 Luna by 12.654.467+12.6
- BrowseComp (context management)GPT-5.6 Luna by 10.873.284+10.8
- SWEBench Pro PublicGPT-5.6 Luna by 10.152.662.7+10.1
- livebench_languageDeepSeek-V4-Flash by 6.679.272.6+6.6
- AA-OmniscienceGPT-5.6 Luna by 4-14.3-10.3+4
- HLE (with tools)GPT-5.6 Luna by 3.845.148.9+3.8
- AA IntelligenceGPT-5.6 Luna by 334.337.3+3
- Toolathlon VerifiedDeepSeek-V4-Flash by 2.470.367.9+2.4
- Artificial Analysis Coding IndexGPT-5.6 Luna by 2.369.171.4+2.3
- ARC-AGI-1GPT-5.6 Luna by 1.78990.7+1.7
- τ³-Bench BankingGPT-5.6 Luna by 1.422.924.3+1.4
- livebench_data_analysisDeepSeek-V4-Flash by 1.379.378+1.3
- AA Agentic IndexGPT-5.6 Luna by 141.742.7+1
Questions people ask
Which is better, DeepSeek-V4-Flash or GPT-5.6 Luna?
GPT-5.6 Luna wins five of the seven areas where both have results: coding, agents, facts, math and long documents. DeepSeek-V4-Flash wins reasoning and following instructions, and costs 71% less.
Which is better for coding?
GPT-5.6 Luna. It wins 7 of the 7 coding tests both models report; DeepSeek-V4-Flash wins none.
Which is cheaper?
DeepSeek-V4-Flash costs $0.14 per million input tokens and $0.28 per million output tokens; GPT-5.6 Luna costs $0.20 and $1.20. That makes DeepSeek-V4-Flash about 71% cheaper for the same work.
How do you compare the two?
We use the 45 benchmark tests both models have published scores on. The verdict counts the 28 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.