GPT-5.6 Luna vs Claude Opus 5
Wins 1 of 8 areas
Long documents
Wins 7 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Following instructions
Claude Opus 5 is the stronger all-rounder.GPT-5.6 Luna is cheaper and better at long documents.
Scores updated · 51 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 thinking07Claude Opus 57 of 7 tests
- CodingWriting and fixing software16Claude Opus 56 of 7 tests
- AgentsCarrying out multi-step tasks on its own05Claude Opus 55 of 5 tests
- FactsGetting facts right instead of making them up03Claude Opus 53 of 3 tests
- Images and chartsUnderstanding pictures, charts and video03Claude Opus 53 of 3 tests
- MathCompetition and research-level math03Claude Opus 53 of 3 tests
- Following instructionsDoing exactly what it is asked01Claude Opus 51 of 1 test
- Long documentsFinding answers in very long texts10GPT-5.6 Luna1 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.
GPT-5.6 Luna costs 95% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Claude Opus 5 pulls ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+37.4points ahead
- Common-sense trick questionsSimpleBench+33.8points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+31.8points ahead
Where GPT-5.6 Luna pulls ahead
- Reasons across sets of long documentsAA-LCR+4.4points ahead
- Programming problems, refreshed regularlyLiveBench · Coding+1.4points ahead
Every test, side by side
All 51 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 5
- LMArena · WebDevClaude Opus 5 by 172 rating points15191691+172 rating
- LiveBench · Agentic CodingClaude Opus 5 by 16.848.465.2+16.8
- SWE-bench ProClaude Opus 5 by 16.562.779.2+16.5
- Terminal-Bench 2.1Claude Opus 5 by 8.280.989.1+8.2
- SWE-bench VerifiedClaude Opus 5 by 39396+3
- SciCodeClaude Opus 5 by 2.853.656.4+2.8
- LiveBench · CodingGPT-5.6 Luna by 1.482.981.5+1.4
AgentsClaude Opus 5
- Terminal-Bench 4.0Claude Opus 5 by 37.411.649+37.4
- GDPValClaude Opus 5 by 13.14861.1+13.1
- τ-Bench V3 · BankingClaude Opus 5 by 1131.142.1+11
- OSWorld-VerifiedClaude Opus 5 by 10.872.683.4+10.8
- BrowseCompClaude Opus 5 by 7.583.390.8+7.5
ReasoningClaude Opus 5
- SimpleBenchClaude Opus 5 by 33.846.880.6+33.8
- ARC-AGI-2Claude Opus 5 by 30.859.690.4+30.8
- ARC-AGI-3Claude Opus 5 by 300.230.2+30
- Humanity's Last ExamClaude Opus 5 by 15.439.554.9+15.4
- CritPtClaude Opus 5 by 8.520.629.1+8.5
- LiveBench · ReasoningClaude Opus 5 by 5.685.691.2+5.6
- GPQA DiamondClaude Opus 5 by 2.191.193.2+2.1
FactsClaude Opus 5
- AA-Omniscience · Non-hallucinationClaude Opus 5 by 31.87.439.2+31.8
- SimpleQA VerifiedClaude Opus 5 by 18.94159.9+18.9
- AA-Omniscience · AccuracyClaude Opus 5 by 18.242.760.9+18.2
Images and chartsClaude Opus 5
- LMArena · VisionClaude Opus 5 by 61 rating points12561317+61 rating
- MMMU-ProClaude Opus 5 by 6.178.684.7+6.1
- CharXiv (RQ)Claude Opus 5 by 182.783.7+1
MathClaude Opus 5
- FrontierMath Tier 4Claude Opus 5 by 14.758.573.2+14.7
- LiveBench · MathematicsClaude Opus 5 by 8.587.295.7+8.5
- FrontierMath Tiers 1-3 (v2)Claude Opus 5 by 3.582.185.6+3.5
Following instructionsClaude Opus 5
- LiveBench · Instruction FollowingClaude Opus 5 by 3.760.163.8+3.7
Other results21 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 Opus 5 by 62.822.785.5+62.8
- AA-OmniscienceClaude Opus 5 by 47.4-10.337.1+47.4
- Terminal-Bench 4.0Claude Opus 5 by 34.517.351.8+34.5
- GDPval-AA v2Claude Opus 5 by 294 rating points15301824+294 rating
- OSWorld 2.0Claude Opus 5 by 2545.670.6+25
- GDPval-AA v2 EloClaude Opus 5 by 240 rating points15841824+240 rating
- Agents' Last ExamGPT-5.6 Luna by 21.750.328.6+21.7
- LVBenchClaude Opus 5 by 16.558.975.4+16.5
- livebench_languageClaude Opus 5 by 16.172.688.7+16.1
- HLE (with tools)Claude Opus 5 by 14.748.963.6+14.7
- AA Agentic IndexClaude Opus 5 by 13.542.756.2+13.5
- AA IntelligenceClaude Opus 5 by 13.537.350.8+13.5
- Toolathlon VerifiedClaude Opus 5 by 12.767.980.6+12.7
- HealthBenchClaude Opus 5 by 11.355.867.1+11.3
- ARC-AGI-1Claude Opus 5 by 6.890.797.5+6.8
- Artificial Analysis Coding IndexClaude Opus 5 by 6.671.478+6.6
- HealthBench ProfessionalClaude Opus 5 by 4.155.759.8+4.1
- HealthBench Professional length-adjustedClaude Opus 5 by 4.155.759.8+4.1
- livebench_data_analysisGPT-5.6 Luna by 3.47874.5+3.4
- HealthBench length-adjustedClaude Opus 5 by 255.857.8+2
- DeepSWE 1.1Claude Opus 5 by 1.86768.8+1.8
Questions people ask
Which is better, GPT-5.6 Luna or Claude Opus 5?
Claude Opus 5 wins seven of the eight areas we test: coding, agents, reasoning, facts, images and charts, math and following instructions. GPT-5.6 Luna wins long documents, and costs 95% less.
Which is better for coding?
Claude Opus 5. It wins 6 of the 7 coding tests both models report; GPT-5.6 Luna wins 1.
Which is cheaper?
GPT-5.6 Luna costs $0.20 per million input tokens and $1.20 per million output tokens; Claude Opus 5 costs $5.00 and $25.00. That makes GPT-5.6 Luna about 95% cheaper for the same work.
How do you compare the two?
We use the 51 benchmark tests both models have published scores on. The verdict counts the 30 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 21 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.