Claude Opus 4.8 vs GPT-5.6 Luna
Wins 6 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Following instructions
Wins 2 of 8 areas
Math · Long documents
Claude Opus 4.8 is the stronger all-rounder.GPT-5.6 Luna is cheaper and better at math.
Scores updated · 52 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 thinking50Claude Opus 4.85 of 7 tests · 2 ties
- AgentsCarrying out multi-step tasks on its own50Claude Opus 4.85 of 6 tests · 1 tie
- FactsGetting facts right instead of making them up30Claude Opus 4.83 of 3 tests
- CodingWriting and fixing software42Claude Opus 4.84 of 7 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video20Claude Opus 4.82 of 2 tests
- Following instructionsDoing exactly what it is asked10Claude Opus 4.81 of 1 test
- MathCompetition and research-level math23GPT-5.6 Luna3 of 5 tests
- Long documentsFinding answers in very long texts01GPT-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 three tests each model wins by the widest margin. Scores are out of 100.
Where Claude Opus 4.8 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+53.3points ahead
- Common-sense trick questionsSimpleBench+18points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+12.5points ahead
Where GPT-5.6 Luna pulls ahead
- Reasons across sets of long documentsAA-LCR+6points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+4.4points ahead
- Harvard-MIT high-school math contest problemsHMMT Feb. 2026+3points ahead
Every test, side by side
All 52 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 4.8
- SWE-bench ProClaude Opus 4.8 by 6.569.262.7+6.5
- SWE-bench VerifiedGPT-5.6 Luna by 4.488.693+4.4
- Terminal-Bench 2.1Claude Opus 4.8 by 3.784.680.9+3.7
- LMArena · WebDevClaude Opus 4.8 by 36 rating points15551519+36 rating
- LiveBench · Agentic CodingClaude Opus 4.8 by 2.150.548.4+2.1
- LiveBench · CodingGPT-5.6 Luna by 1.181.882.9+1.1
- SciCodetie54.453.6tie
AgentsClaude Opus 4.8
- OSWorld-VerifiedClaude Opus 4.8 by 10.883.472.6+10.8
- Terminal-Bench 4.0Claude Opus 4.8 by 10.121.711.6+10.1
- AA ApexAgentsClaude Opus 4.8 by 3.639.435.8+3.6
- τ-Bench V3 · BankingClaude Opus 4.8 by 3.134.231.1+3.1
- BrowseCompClaude Opus 4.8 by 184.383.3+1
- GDPValtie47.748tie
ReasoningClaude Opus 4.8
- SimpleBenchClaude Opus 4.8 by 1864.846.8+18
- ARC-AGI-2Claude Opus 4.8 by 12.572.159.6+12.5
- Humanity's Last ExamClaude Opus 4.8 by 9.248.739.5+9.2
- LiveBench · ReasoningClaude Opus 4.8 by 3.689.285.6+3.6
- ARC-AGI-3Claude Opus 4.8 by 1.31.50.2+1.3
- GPQA Diamondtie9291.1tie
- CritPttie20.920.6tie
FactsClaude Opus 4.8
- AA-Omniscience · Non-hallucinationClaude Opus 4.8 by 53.360.77.4+53.3
- SimpleQA VerifiedClaude Opus 4.8 by 125341+12
- AA-Omniscience · AccuracyClaude Opus 4.8 by 6.148.842.7+6.1
Images and chartsClaude Opus 4.8
- CharXiv (RQ)Claude Opus 4.8 by 7.289.982.7+7.2
- LMArena · VisionClaude Opus 4.8 by 39 rating points12951256+39 rating
MathGPT-5.6 Luna
- LiveBench · MathematicsClaude Opus 4.8 by 7.194.387.2+7.1
- HMMT Feb. 2026GPT-5.6 Luna by 395.598.5+3
- AIME 2026Claude Opus 4.8 by 2.410097.6+2.4
- FrontierMath Tier 4GPT-5.6 Luna by 2.456.158.5+2.4
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Luna by 2.18082.1+2.1
Following instructionsClaude Opus 4.8
- LiveBench · Instruction FollowingClaude Opus 4.8 by 11.97260.1+11.9
Other results20 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 4.8 by 62.184.822.7+62.1
- AA-OmniscienceClaude Opus 4.8 by 39.128.8-10.3+39.1
- Agents' Last ExamGPT-5.6 Luna by 24.625.750.3+24.6
- livebench_data_analysisGPT-5.6 Luna by 126678+12
- Toolathlon VerifiedClaude Opus 4.8 by 1279.967.9+12
- OSWorld 2.0Claude Opus 4.8 by 10.155.745.6+10.1
- DeepSWEGPT-5.6 Luna by 9.25867.2+9.2
- HLE (with tools)Claude Opus 4.8 by 957.948.9+9
- DeepSWE 1.1GPT-5.6 Luna by 85967+8
- livebench_languageClaude Opus 4.8 by 7.179.772.6+7.1
- ToolathlonClaude Opus 4.8 by 6.559.953.4+6.5
- GDPval-AA v2Claude Opus 4.8 by 63 rating points15931530+63 rating
- Terminal-Bench 4.0Claude Opus 4.8 by 6.323.617.3+6.3
- AA IntelligenceClaude Opus 4.8 by 4.541.837.3+4.5
- HealthBenchClaude Opus 4.8 by 3.559.355.8+3.5
- Artificial Analysis Coding IndexClaude Opus 4.8 by 2.974.371.4+2.9
- ARC-AGI-1Claude Opus 4.8 by 1.892.590.7+1.8
- GDPval-AA v2 Elotie15931584tie
- AA Agentic Indextie42.642.7tie
- HealthBench Professionaltie55.855.7tie
Questions people ask
Which is better, Claude Opus 4.8 or GPT-5.6 Luna?
Claude Opus 4.8 wins six of the eight areas we test: coding, agents, reasoning, facts, images and charts and following instructions. GPT-5.6 Luna wins math and long documents, and costs 95% less.
Which is better for coding?
Claude Opus 4.8. It wins 4 of the 7 coding tests both models report; GPT-5.6 Luna wins 2, and 1 is a tie.
Which is cheaper?
Claude Opus 4.8 costs $5.00 per million input tokens and $25.00 per million output tokens; GPT-5.6 Luna costs $0.20 and $1.20. That makes GPT-5.6 Luna about 95% cheaper for the same work.
How do you compare the two?
We use the 52 benchmark tests both models have published scores on. The verdict counts the 32 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 20 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.