Up to four models
GPT-6 Luna vs DeepSeek-V4.1-Flash
On the index (GPT-4 = 100), GPT-6 Luna scores 170 and DeepSeek-V4.1-Flash 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-6 Luna
170153–186
WeightsClosed
ReleasedSep 2026
Price in / out$0.10 / $0.50
Context1.05M
Core tests taken6 of 9
DeepSeek-V4.1-Flash170153–186
WeightsOpen
ReleasedSep 2026
Price in / out$0.14 / $0.42
Context1.05M
Core tests taken6 of 9
GPT-6 Luna and DeepSeek-V4.1-Flash are too close to call (0 points apart, within the uncertainty), ahead on 3 of 6 shared core tests, at a similar price.
Test by testthird-party results
GPT-6 LunaDeepSeek-V4.1-Flash
DeepSeek-V4.1-FlashEvery result as a table
| Test | GPT-6 Luna | DeepSeek-V4.1-Flash |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 90.5% | 89.8% |
| LiveBench Data Analysis | 73.4% | 79.3% |
| LiveBench IF | 55.9% | 70.0% |
| LiveBench Language | 73.8% | 81.2% |
| LiveBench Reasoning | 81.8% | 86.7% |
| SimpleQA Verified | 41.4% | 42.9% |
| Maths | ||
| FrontierMath (tiers 1-3) | 79.0% | 67.4% |
| AIME-style maths (OTIS mock) | 98.9% | 98.3% |
| LiveBench Mathematics | 89.1% | 93.3% |
| ProofBench v1.1 | 64.0% | 54.0% |
| Chess puzzles | 36.0% | 48.0% |
| FrontierMath (tier 4) | 56.1% | 26.8% |
| Coding | ||
| Terminal-Bench 4.0 | 13.6% | 19.7% |
| LiveBench Agentic Coding | 51.2% | 77.3% |
| LiveBench Coding | 79.0% | 80.0% |
| Code Migration | 42.5% | 45.6% |
| CyberBench v1.1 | 76.3% | 73.7% |
| IOI | 55.6% | 40.3% |
| ProgramBench | 0.5% | 0.5% |
| Vibe Code Bench v1.1 | 81.7% | 84.7% |
| Agents | ||
| APEX-Agents | 44.3% | 39.5% |
| BioMysteryBench | 61.5% | 67.8% |
| SRE Bench | 2.7% | 0.8% |
| Terminal-Bench-Science | 4.3% | 0.0% |
| DTBench | 90.1% | 89.9% |
| Furniture assembly | 44.2% | 34.2% |
| LMCA | 44.5% | 47.0% |
| Mystery game puzzles | 7.0% | 43.0% |
| Professional work | ||
| Finance Agent (v2) | 49.9% | 53.5% |
| EMB | 68.5% | 57.2% |
| Legal Research Bench | 30.3% | 41.3% |
| MedCode | 44.7% | 41.2% |
| MedScribe | 83.7% | 85.5% |
| Public Benefits Bench v1.1 | 57.6% | 64.3% |
| SAGE | 48.1% | 47.9% |
| Tax Agent Bench | 20.0% | 26.6% |
| Novel problems | ||
| ARC-AGI-2 | 59.3% | 72.9% |
| ARC-AGI-1 | 86.7% | 94.5% |
| ARC-AGI-3 | 0.2% | 0.9% |
| MysteryMechanism | 19.4% | 21.2% |
Tests at least two of these models have taken. Bold rows are the core tests.
