Up to four models
GPT-5.6 Luna vs GPT-6 Luna
On the index (GPT-4 = 100), GPT-5.6 Luna scores 170 and GPT-6 Luna 170. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
GPT-5.6 Luna
170154–187
WeightsClosed
ReleasedJul 2026
Price in / out$0.20 / $1.2
Context1.05M
Core tests taken6 of 9
GPT-6 Luna
170153–186
WeightsClosed
ReleasedSep 2026
Price in / out$0.10 / $0.50
Context1.05M
Core tests taken6 of 9
GPT-5.6 Luna and GPT-6 Luna are too close to call (1 point apart, within the uncertainty), ahead on 4 of 6 shared core tests, at 2.3× the price.
Test by testthird-party results
GPT-5.6 LunaGPT-6 Luna
Every result as a table
| Test | GPT-5.6 Luna | GPT-6 Luna |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 91.6% | 90.5% |
| LiveBench Data Analysis | 78.0% | 73.4% |
| LiveBench IF | 60.1% | 55.9% |
| LiveBench Language | 72.6% | 73.8% |
| LiveBench Reasoning | 85.6% | 81.8% |
| SimpleQA Verified | 41.0% | 41.4% |
| Maths | ||
| FrontierMath (tiers 1-3) | 82.1% | 79.0% |
| AIME-style maths (OTIS mock) | 98.3% | 98.9% |
| LiveBench Mathematics | 87.2% | 89.1% |
| ProofBench v1.1 | 60.0% | 64.0% |
| Chess puzzles | 43.0% | 36.0% |
| FrontierMath (tier 4) | 61.0% | 56.1% |
| Coding | ||
| Terminal-Bench 4.0 | 11.6% | 13.6% |
| LiveBench Agentic Coding | 48.4% | 51.2% |
| LiveBench Coding | 82.9% | 79.0% |
| Code Migration | 44.5% | 42.5% |
| CyberBench v1.1 | 73.6% | 76.3% |
| IOI | 61.8% | 55.6% |
| ProgramBench | 0.0% | 0.5% |
| Vibe Code Bench v1.1 | 77.1% | 81.7% |
| Agents | ||
| APEX-Agents | 43.0% | 44.3% |
| BioMysteryBench | 61.5% | 61.5% |
| SRE Bench | 1.5% | 2.7% |
| Terminal-Bench-Science | 0.0% | 4.3% |
| DTBench | 89.1% | 90.1% |
| Furniture assembly | 42.5% | 44.2% |
| LMCA | 48.5% | 44.5% |
| Mystery game puzzles | 21.0% | 7.0% |
| Professional work | ||
| Finance Agent (v2) | 55.0% | 49.9% |
| EMB | 67.1% | 68.5% |
| Legal Research Bench | 36.5% | 30.3% |
| MedCode | 42.4% | 44.7% |
| MedScribe | 84.4% | 83.7% |
| Public Benefits Bench v1.1 | 61.2% | 57.6% |
| SAGE | 44.2% | 48.1% |
| Tax Agent Bench | 24.8% | 20.0% |
| Novel problems | ||
| ARC-AGI-2 | 59.6% | 59.3% |
| ARC-AGI-1 | 90.7% | 86.7% |
| ARC-AGI-3 | 0.2% | 0.2% |
| MysteryMechanism | 14.4% | 19.4% |
Tests at least two of these models have taken. Bold rows are the core tests.
