Up to four models
GPT-5.6 Luna vs Gemini 3.1 Pro
On the index (GPT-4 = 100), GPT-5.6 Luna scores 162 and Gemini 3.1 Pro 161. 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
162146–178
WeightsClosed
ReleasedJul 2026
Price in / out$0.20 / $1.2
Context1.05M
Core tests taken6 of 9
161146–177
WeightsClosed
ReleasedFeb 2026
Price in / out$2.0 / $12
Context1.05M
Core tests taken9 of 9
GPT-5.6 Luna and Gemini 3.1 Pro are too close to call (1 point apart, within the uncertainty), ahead on 4 of 6 shared core tests, and costs 10× less.
Test by testthird-party results
GPT-5.6 Luna
Gemini 3.1 Pro
Every result as a table
| Test | GPT-5.6 Luna | Gemini 3.1 Pro |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 91.6% | 94.4% |
| GPQA Diamond (Vals AI) | 91.7% | 95.5% |
| MMLU Pro | 86.0% | 91.0% |
| MMMU Pro | 85.0% | 88.2% |
| SimpleBench | 46.8% | 79.6% |
| SimpleQA Verified | 41.0% | 73.5% |
| Maths | ||
| FrontierMath (tiers 1-3) | 82.1% | 59.7% |
| AIME-style maths (OTIS mock) | 98.3% | 95.6% |
| ProofBench v1.1 | 60.0% | 26.0% |
| Chess puzzles | 43.0% | 51.0% |
| FrontierMath (tier 4) | 61.0% | 26.8% |
| Coding | ||
| Terminal-Bench 4.0 | 11.6% | 2.5% |
| Code Migration | 44.5% | 17.3% |
| IOI | 61.8% | 51.8% |
| ProgramBench | 0.0% | 0.0% |
| SWE-bench | 93.0% | 78.8% |
| Vibe Code Bench 1-100 | 22.6% | 6.7% |
| Vibe Code Bench v1.1 | 77.1% | 32.0% |
| DeepSWE | 67.2% | 11.7% |
| MirrorCode | 21.1% | 8.9% |
| WeirdML | 60.9% | 72.1% |
| Agents | ||
| APEX-Agents | 43.0% | 35.1% |
| BioMysteryBench | 61.5% | 60.0% |
| SRE Bench | 1.5% | 0.0% |
| Terminal-Bench-Science | 0.0% | 1.4% |
| BALROG | 45.6% | 57.0% |
| DTBench | 89.1% | 97.1% |
| Furniture assembly | 42.5% | 26.7% |
| LMCA | 48.5% | 53.8% |
| Mystery game puzzles | 21.0% | 34.0% |
| Professional work | ||
| Finance Agent (v2) | 55.0% | 43.0% |
| CorpFin v2 | 64.2% | 64.5% |
| EMB | 67.1% | 52.6% |
| LegalBench | 84.0% | 87.4% |
| Legal Research Bench | 36.5% | 20.7% |
| MedCode | 42.4% | 59.1% |
| MedScribe | 84.4% | 76.1% |
| MortgageTax | 67.3% | 69.4% |
| Public Benefits Bench v1.1 | 61.2% | 53.8% |
| Tax Agent Bench | 24.8% | 9.3% |
| TaxEval v2 | 76.2% | 72.9% |
| Novel problems | ||
| ARC-AGI-2 | 59.6% | 77.1% |
| ARC-AGI-1 | 90.7% | 98.0% |
| ARC-AGI-3 | 0.2% | 0.4% |
| MysteryMechanism | 14.4% | 20.7% |
Task length · METR
46 sec
Tests at least two of these models have taken. Bold rows are the core tests.