GPT-5.4 vs GPT-5.6 Luna
Wins 4 of 8 areas
Reasoning · Facts · Images and charts · Following instructions
Wins 3 of 8 areas
Coding · Agents · Long documents
GPT-5.4 wins more areas, narrowly.GPT-5.6 Luna is cheaper and better at coding.
Scores updated · 46 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 thinking40GPT-5.44 of 6 tests · 2 ties
- FactsGetting facts right instead of making them up20GPT-5.42 of 3 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video10GPT-5.41 of 3 tests · 2 ties
- Following instructionsDoing exactly what it is asked10GPT-5.41 of 1 test
- CodingWriting and fixing software24GPT-5.6 Luna4 of 6 tests
- AgentsCarrying out multi-step tasks on its own23GPT-5.6 Luna3 of 6 tests · 1 tie
- Long documentsFinding answers in very long texts01GPT-5.6 Luna1 of 1 test
- MathCompetition and research-level math22Even2 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 92% 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
- Abstract visual puzzles that people can solveARC-AGI-2+13.7points ahead
- Follows detailed instructions exactlyLiveBench · Instruction Following+10.1points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+8.5points ahead
Where GPT-5.6 Luna pulls ahead
- Real work tasks from 44 professionsGDPVal+10.7points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+9.5points ahead
- Finds the root cause of IT system incidentsAA IT-Bench SRE+5.8points 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-5.6 Luna
- LMArena · WebDevGPT-5.6 Luna by 57 rating points14621519+57 rating
- LiveBench · Agentic CodingGPT-5.4 by 5.453.848.4+5.4
- LiveBench · CodingGPT-5.6 Luna by 5.477.582.9+5.4
- SWE-bench ProGPT-5.6 Luna by 557.762.7+5
- SciCodeGPT-5.4 by 356.653.6+3
- Terminal-Bench 2.1GPT-5.6 Luna by 2.678.380.9+2.6
AgentsGPT-5.6 Luna
- GDPValGPT-5.6 Luna by 10.737.348+10.7
- τ-Bench V3 · BankingGPT-5.4 by 8.539.631.1+8.5
- AA IT-Bench SREGPT-5.6 Luna by 5.834.540.3+5.8
- AA ApexAgentsGPT-5.6 Luna by 2.533.335.8+2.5
- OSWorld-VerifiedGPT-5.4 by 2.47572.6+2.4
- BrowseComptie82.783.3tie
ReasoningGPT-5.4
- ARC-AGI-2GPT-5.4 by 13.773.359.6+13.7
- Humanity's Last ExamGPT-5.4 by 4.243.739.5+4.2
- CritPtGPT-5.4 by 2.823.420.6+2.8
- LiveBench · ReasoningGPT-5.4 by 2.588.185.6+2.5
- GPQA Diamondtie9291.1tie
- ARC-AGI-3tie0.20.2tie
FactsGPT-5.4
- AA-Omniscience · AccuracyGPT-5.4 by 8.250.942.7+8.2
- SimpleQA VerifiedGPT-5.4 by 4.145.141+4.1
- AA-Omniscience · Non-hallucinationtie8.37.4tie
Images and chartsGPT-5.4
- LMArena · VisionGPT-5.4 by 45 rating points13011256+45 rating
- MMMU-Protie78.478.6tie
- CharXiv (RQ)tie82.882.7tie
MathEven
- FrontierMath Tier 4GPT-5.6 Luna by 9.54958.5+9.5
- LiveBench · MathematicsGPT-5.4 by 794.287.2+7
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Luna by 3.578.682.1+3.5
- AIME 2026GPT-5.4 by 1.699.297.6+1.6
- HMMT Feb. 2026tie97.798.5tie
Following instructionsGPT-5.4
- LiveBench · Instruction FollowingGPT-5.4 by 10.170.260.1+10.1
Other results15 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.
- FrontierMath (overall)GPT-5.6 Luna by 3147.678.6+31
- AA-OmniscienceGPT-5.4 by 16.15.8-10.3+16.1
- DeepSWE 1.1GPT-5.6 Luna by 155267+15
- livebench_languageGPT-5.4 by 1082.672.6+10
- HealthBench Professional length-adjustedGPT-5.6 Luna by 7.648.155.7+7.6
- HLE (with tools)GPT-5.4 by 3.252.148.9+3.2
- ARC-AGI-1GPT-5.4 by 393.790.7+3
- HealthBench Hard length-adjustedGPT-5.6 Luna by 2.929.132+2.9
- HealthBench length-adjustedGPT-5.6 Luna by 1.85455.8+1.8
- AA IntelligenceGPT-5.4 by 1.73937.3+1.7
- AA Agentic IndexGPT-5.4 by 1.544.242.7+1.5
- livebench_data_analysisGPT-5.4 by 1.379.378+1.3
- HealthBench Consensus length-adjustedGPT-5.4 by 1.296.395.1+1.2
- ToolathlonGPT-5.4 by 1.254.653.4+1.2
- Artificial Analysis Coding Indextie71.171.4tie
Questions people ask
Which is better, GPT-5.4 or GPT-5.6 Luna?
GPT-5.4 wins four of the eight areas we test: reasoning, facts, images and charts and following instructions. GPT-5.6 Luna wins coding, agents and long documents, and costs 92% less. They are level on math.
Which is better for coding?
GPT-5.6 Luna. It wins 4 of the 6 coding tests both models report; GPT-5.4 wins 2.
Which is cheaper?
GPT-5.4 costs $2.50 per million input tokens and $15.00 per million output tokens; GPT-5.6 Luna costs $0.20 and $1.20. That makes GPT-5.6 Luna about 92% 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 31 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 15 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.