OpenAI·released 2026-09-221 source
GPT-6 Luna benchmark scores: 16 benchmarks tracked. 0% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 16 benchmarks·best result 98.89 on OTIS Mock AIME 2024-2025 (unverified)·0 of 16 independently reproduced
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.89accuracy (%) | unverifiedT2 | 2026-09-22 |
| GPQA diamond | 87.29accuracy (%) | unverifiedT2 | 2026-09-22 |
| ARC-AGI | 86.67accuracy (%) | unverifiedT1 | 2026-09-22https://arcprize.org/leaderboard |
| DTBench | 83.55accuracy (%) | unverifiedT1 | 2026-09-22https://conceptualreasoning.ai/dtbench |
| FrontierMath-Tiers-1-3-v2-Private | 78.95accuracy (%) | unverifiedT2 | 2026-09-22 |
| ProofBench | 64accuracy (%) | unverifiedT1 | 2026-09-22https://www.vals.ai/benchmarks/proof_bench |
| ARC-AGI-2 | 59.31accuracy (%) | unverifiedT2 | 2026-09-22 |
| FrontierMath-Tier-4-v2-Private | 56.1accuracy (%) | unverifiedT2 | 2026-09-22 |
| LMCA | 52.34accuracy (%) | unverifiedT1 | 2026-09-22https://conceptualreasoning.ai/lmca |
| APEX-Agents | 44.3accuracy (%) | unverifiedT2 | 2026-09-22 |
| FrontierCode | 42.42accuracy (%) | unverifiedT1 | 2026-09-22https://cognition.com/frontiercode |
| SimpleQA Verified | 41.4accuracy (%) | unverifiedT2 | 2026-09-22 |
| Chess Puzzles | 32.66accuracy (%) | unverifiedT2 | 2026-09-22 |
| Furniture Assembly | 20.24accuracy (%) | unverifiedT2 | 2026-09-22 |
| Mystery Game Puzzles | 0accuracy (%) | unverifiedT2 | 2026-09-22 |
| SpireClimb | 0accuracy (%) | unverifiedT2 | 2026-09-22 |
Claims drawn from cited facts, not live model generation.
This model comes from OpenAI and was released in 2026.
2 cited facts
This model is scored on 16 tracked benchmarks and holds no current top score. Its benchmark record is unverified, meaning outside evaluation has neither confirmed nor disputed the reported numbers.
3 cited facts
On average this model trails the leading score by 30.17 points across its benchmarks, a score gap measured under a disclosed harness rather than a claim about capability. A gap of that width reads as well behind the leaders rather than at the frontier, and it holds the top score on none of the benchmarks it reports, meaning no category leadership yet.
3 cited facts
1 facts cross-checked across data sources: 1 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks
GPT-6 Luna is an AI model developed by OpenAI, released 2026-09-22. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
GPT-6 Luna has recorded scores on 16 benchmarks, each shown with its evidence status.
GPT-6 Luna has recorded scores on 16 benchmarks — GPQA diamond, OTIS Mock AIME 2024-2025, DTBench, LMCA, SimpleQA Verified, Chess Puzzles, and 10 more. The full table above shows each score with its evidence status.
0 of 16 recorded scores (0%) are independently reproduced rather than self-reported by the lab.
GPT-6 Luna has 16 tracked claims: 16 unverified.