GPT-5.6 Luna (Max): benchmark results
GPT-5.6 Luna evaluated at the max reasoning-effort setting. Provider: OpenAI. Released 2026-07-09. Access: API.
Aggregate Rating 1802 ± 8, rank #135 of 2168 rated models, from 296 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| IFM K2-Horizon-375B Card - SWE-bench Pro (Strict, No Internet) | 48.8 | Resolved (%) | 100 |
| Nejumi 4 - JBBQ (2-shot) - Accuracy | 96 | Accuracy (%) | 99 |
| Horangi 4 - HAE-RAE Bench (Reading Comprehension) | 93 | Accuracy (%) | 98.6 |
| Nejumi 4 - JTruthfulQA | 77.15 | Score (%) | 98.6 |
| AA Long Context Reasoning | 83.67 | Accuracy (%) | 97.2 |
| LiveBench Code Completion | 86.96 | Score | 97.1 |
| Nejumi 4 - jaster (2-shot) - MAWPS | 99 | Exact match (%) | 96.9 |
| Horangi 4 - HumanEval (100-problem subset) | 100 | Pass rate (%) | 96.2 |
| Vals AI TaxEval v2 | 76.17 | Accuracy (%) | 96.2 |
| Korean CSAT 2026 (Easy Mode) - Physics I | 50 | Points (out of 50) | 95.9 |
| IOI | 72.92 | Score (imported) | 95 |
| Horangi 4 - HAE-RAE Bench (without Reading Comprehension) | 98.99 | Accuracy (%) | 94.7 |
Interactive version: theaggregate.ai/model?slug=gpt-5-6-luna-max · How It Works · Data refreshed daily, snapshot 2026-10-20.