Claude Sonnet 5 vs GPT-5.6 Luna
Wins 4 of 8 areas
Agents · Reasoning · Images and charts · Following instructions
Wins 2 of 8 areas
Facts · Math
Claude Sonnet 5 wins more areas, narrowly.GPT-5.6 Luna is cheaper and better at facts.
Scores updated · 47 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 own40Claude Sonnet 54 of 5 tests · 1 tie
- ReasoningHard problems that need careful thinking31Claude Sonnet 53 of 5 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video21Claude Sonnet 52 of 3 tests
- Following instructionsDoing exactly what it is asked10Claude Sonnet 51 of 1 test
- FactsGetting facts right instead of making them up12GPT-5.6 Luna2 of 3 tests
- MathCompetition and research-level math12GPT-5.6 Luna2 of 3 tests
- CodingWriting and fixing software22Even2 each · 3 ties
- Long documentsFinding answers in very long texts11Even1 each
Following instructions 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.
GPT-5.6 Luna costs 88% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Claude Sonnet 5 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+53.2points ahead
- Common-sense trick questionsSimpleBench+13.8points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+11points ahead
Where GPT-5.6 Luna pulls ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+29.2points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+16.5points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+7.8points ahead
Every test, side by side
All 47 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingEven
- LiveBench · Agentic CodingClaude Sonnet 5 by 1159.448.4+11
- SWE-bench VerifiedGPT-5.6 Luna by 7.885.293+7.8
- LiveBench · CodingGPT-5.6 Luna by 2.280.782.9+2.2
- LMArena · WebDevClaude Sonnet 5 by 20 rating points15391519+20 rating
- SciCodetie54.353.6tie
- SWE-bench Protie63.262.7tie
- Terminal-Bench 2.1tie80.580.9tie
AgentsClaude Sonnet 5
- OSWorld-VerifiedClaude Sonnet 5 by 8.681.272.6+8.6
- τ-Bench V3 · BankingClaude Sonnet 5 by 6.237.331.1+6.2
- Terminal-Bench 4.0Claude Sonnet 5 by 2.514.111.6+2.5
- BrowseCompClaude Sonnet 5 by 1.484.783.3+1.4
- GDPValtie48.248tie
ReasoningClaude Sonnet 5
- SimpleBenchClaude Sonnet 5 by 13.860.646.8+13.8
- CritPtGPT-5.6 Luna by 3.716.920.6+3.7
- LiveBench · ReasoningClaude Sonnet 5 by 3.188.785.6+3.1
- Humanity's Last ExamClaude Sonnet 5 by 1.841.339.5+1.8
- GPQA Diamondtie91.191.1tie
FactsGPT-5.6 Luna
- AA-Omniscience · Non-hallucinationClaude Sonnet 5 by 53.260.67.4+53.2
- SimpleQA VerifiedGPT-5.6 Luna by 7.333.741+7.3
- AA-Omniscience · AccuracyGPT-5.6 Luna by 2.64042.7+2.6
Images and chartsClaude Sonnet 5
- CharXiv (RQ)Claude Sonnet 5 by 5.688.382.7+5.6
- LMArena · VisionClaude Sonnet 5 by 16 rating points12721256+16 rating
- MMMU-ProGPT-5.6 Luna by 1.377.378.6+1.3
MathGPT-5.6 Luna
- FrontierMath Tier 4GPT-5.6 Luna by 29.229.358.5+29.2
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Luna by 16.565.682.1+16.5
- LiveBench · MathematicsClaude Sonnet 5 by 5.792.987.2+5.7
Long documentsEven
- MRCR v2 8 needle 128k (average)Claude Sonnet 5 by 6.781.574.8+6.7
- AA-LCRGPT-5.6 Luna by 1.78283.7+1.7
Following instructionsClaude Sonnet 5
- LiveBench · Instruction FollowingClaude Sonnet 5 by 3.863.960.1+3.8
Other results18 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.
- GDP (Surge AI)Claude Sonnet 5 by 58.981.622.7+58.9
- AA-OmniscienceClaude Sonnet 5 by 26.816.4-10.3+26.8
- MLE-BenchClaude Sonnet 5 by 19.366.947.6+19.3
- DeepSWE 1.1GPT-5.6 Luna by 135467+13
- LVBenchClaude Sonnet 5 by 9.668.558.9+9.6
- GDPval-AA v2Claude Sonnet 5 by 88 rating points16181530+88 rating
- HLE (with tools)Claude Sonnet 5 by 8.557.448.9+8.5
- Toolathlon VerifiedClaude Sonnet 5 by 6.874.767.9+6.8
- livebench_data_analysisGPT-5.6 Luna by 6.371.778+6.3
- Terminal-Bench 4.0GPT-5.6 Luna by 4.912.417.3+4.9
- HealthBenchClaude Sonnet 5 by 3.459.255.8+3.4
- livebench_languageClaude Sonnet 5 by 2.47572.6+2.4
- HealthBench ProfessionalClaude Sonnet 5 by 2.157.855.7+2.1
- AA Agentic IndexClaude Sonnet 5 by 1.644.342.7+1.6
- AA Intelligencetie38.237.3tie
- Toolathlontie54.353.4tie
- Artificial Analysis Coding Indextie71.571.4tie
- GDPval-AA v2 Elotie15841584tie
Questions people ask
Which is better, Claude Sonnet 5 or GPT-5.6 Luna?
Claude Sonnet 5 wins four of the eight areas we test: agents, reasoning, images and charts and following instructions. GPT-5.6 Luna wins facts and math, and costs 88% less. They are level on coding and long documents.
Which is better for coding?
Neither. They win 2 coding tests each of the 7 both models report, and 3 are ties.
Which is cheaper?
Claude Sonnet 5 costs $2.00 per million input tokens and $10.00 per million output tokens; GPT-5.6 Luna costs $0.20 and $1.20. That makes GPT-5.6 Luna about 88% cheaper for the same work.
How do you compare the two?
We use the 47 benchmark tests both models have published scores on. The verdict counts the 29 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 18 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.