GPT-5.6 Luna vs GPT-5.6 Sol
Wins 0 of 8 areas
—
Wins 7 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Following instructions
GPT-5.6 Sol is the stronger all-rounder.GPT-5.6 Luna is cheaper.
Scores updated · 61 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 own07GPT-5.6 Sol7 of 7 tests
- ReasoningHard problems that need careful thinking07GPT-5.6 Sol7 of 7 tests
- CodingWriting and fixing software06GPT-5.6 Sol6 of 6 tests
- MathCompetition and research-level math04GPT-5.6 Sol4 of 4 tests
- Images and chartsUnderstanding pictures, charts and video03GPT-5.6 Sol3 of 3 tests
- FactsGetting facts right instead of making them up02GPT-5.6 Sol2 of 3 tests · 1 tie
- Following instructionsDoing exactly what it is asked01GPT-5.6 Sol1 of 1 test
- Long documentsFinding answers in very long texts00Even0 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 94% 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 GPT-5.6 Sol pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+32.9points ahead
- Short factual questions, answered correctlySimpleQA Verified+28.7points ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+28.3points ahead
Where GPT-5.6 Luna pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 61 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.6 Sol
- LMArena · WebDevGPT-5.6 Sol by 96 rating points15191615+96 rating
- LiveBench · Agentic CodingGPT-5.6 Sol by 7.848.456.2+7.8
- Terminal-Bench 2.1GPT-5.6 Sol by 7.180.988+7.1
- SciCodeGPT-5.6 Sol by 3.553.657.1+3.5
- SWE-bench ProGPT-5.6 Sol by 1.962.764.6+1.9
- LiveBench · CodingGPT-5.6 Sol by 182.983.9+1
AgentsGPT-5.6 Sol
- Terminal-Bench 4.0GPT-5.6 Sol by 28.311.639.9+28.3
- AA IT-Bench SREGPT-5.6 Sol by 15.940.356.2+15.9
- τ-Bench V3 · BankingGPT-5.6 Sol by 13.231.144.3+13.2
- OSWorld-VerifiedGPT-5.6 Sol by 10.472.683+10.4
- GDPValGPT-5.6 Sol by 7.44855.4+7.4
- BrowseCompGPT-5.6 Sol by 7.183.390.4+7.1
- AA ApexAgentsGPT-5.6 Sol by 4.135.839.9+4.1
ReasoningGPT-5.6 Sol
- ARC-AGI-2GPT-5.6 Sol by 32.959.692.5+32.9
- SimpleBenchGPT-5.6 Sol by 1846.864.8+18
- CritPtGPT-5.6 Sol by 11.720.632.3+11.7
- Humanity's Last ExamGPT-5.6 Sol by 1039.549.5+10
- ARC-AGI-3GPT-5.6 Sol by 7.60.27.8+7.6
- LiveBench · ReasoningGPT-5.6 Sol by 6.185.691.7+6.1
- GPQA DiamondGPT-5.6 Sol by 391.194.1+3
FactsGPT-5.6 Sol
- SimpleQA VerifiedGPT-5.6 Sol by 28.74169.7+28.7
- AA-Omniscience · AccuracyGPT-5.6 Sol by 16.742.759.4+16.7
- AA-Omniscience · Non-hallucinationtie7.47.8tie
Images and chartsGPT-5.6 Sol
- MMMU-ProGPT-5.6 Sol by 4.878.683.4+4.8
- CharXiv (RQ)GPT-5.6 Sol by 3.182.785.8+3.1
- LMArena · VisionGPT-5.6 Sol by 26 rating points12561282+26 rating
MathGPT-5.6 Sol
- FrontierMath Tier 4GPT-5.6 Sol by 24.558.583+24.5
- LiveBench · MathematicsGPT-5.6 Sol by 987.296.2+9
- FrontierMath Tiers 1-3 (v2)GPT-5.6 Sol by 782.189.1+7
- AIME 2026GPT-5.6 Sol by 2.397.699.9+2.3
Following instructionsGPT-5.6 Sol
- LiveBench · Instruction FollowingGPT-5.6 Sol by 11.860.171.8+11.8
Other results29 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.
- AA-OmniscienceGPT-5.6 Sol by 32.3-10.322+32.3
- LVBenchGPT-5.6 Sol by 23.258.982.1+23.2
- Terminal-Bench 4.0GPT-5.6 Sol by 2017.337.3+20
- GDPval-AA v2GPT-5.6 Sol by 181 rating points15301711+181 rating
- OSWorld 2.0GPT-5.6 Sol by 1745.662.6+17
- HLE (with tools)GPT-5.6 Sol by 15.648.964.5+15.6
- livebench_languageGPT-5.6 Sol by 15.172.687.7+15.1
- GDPval-AA v2 EloGPT-5.6 Sol by 126 rating points15841710+126 rating
- FrontierMath (overall)GPT-5.6 Sol by 10.478.689+10.4
- AA IntelligenceGPT-5.6 Sol by 9.737.347+9.7
- τ³-Bench BankingGPT-5.6 Sol by 8.724.333+8.7
- GDP (Surge AI)GPT-5.6 Sol by 822.730.7+8
- AA Agentic IndexGPT-5.6 Sol by 7.842.750.5+7.8
- Toolathlon VerifiedGPT-5.6 Sol by 767.974.9+7
- Artificial Analysis Coding IndexGPT-5.6 Sol by 671.477.4+6
- DeepSWE 1.1GPT-5.6 Sol by 66773+6
- ARC-AGI-1GPT-5.6 Sol by 5.890.796.5+5.8
- DeepSWEGPT-5.6 Sol by 5.567.272.7+5.5
- HealthBench ProfessionalGPT-5.6 Sol by 4.855.760.5+4.8
- HealthBench Professional length-adjustedGPT-5.6 Sol by 4.855.760.5+4.8
- ToolathlonGPT-5.6 Sol by 4.653.458+4.6
- Agents' Last ExamGPT-5.6 Sol by 2.450.352.7+2.4
- SWEBench Pro PublicGPT-5.6 Sol by 1.962.764.6+1.9
- livebench_data_analysisGPT-5.6 Sol by 1.87879.8+1.8
- HealthBenchGPT-5.6 Sol by 1.255.857+1.2
- HealthBench length-adjustedGPT-5.6 Sol by 1.255.857+1.2
- HealthBench HardGPT-5.6 Sol by 1.13233.1+1.1
- HealthBench Hard length-adjustedGPT-5.6 Sol by 1.13233.1+1.1
- HealthBench Consensus length-adjustedtie95.195.5tie
Questions people ask
Which is better, GPT-5.6 Luna or GPT-5.6 Sol?
GPT-5.6 Sol wins seven of the eight areas we test: coding, agents, reasoning, facts, images and charts, math and following instructions. GPT-5.6 Luna wins none, but costs 94% less. They are level on long documents.
Which is better for coding?
GPT-5.6 Sol. It wins 6 of the 6 coding tests both models report; GPT-5.6 Luna wins none.
Which is cheaper?
GPT-5.6 Luna costs $0.20 per million input tokens and $1.20 per million output tokens; GPT-5.6 Sol costs $4.00 and $20.00. That makes GPT-5.6 Luna about 94% cheaper for the same work.
How do you compare the two?
We use the 61 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 29 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.