GPT-5.6 Luna: benchmark results
OpenAI's fast, low-cost GPT-5.6 tier. Provider: OpenAI. Released 2026-07-09. Access: API.
Aggregate Rating 1797 ± 4, rank #68 of 1656 rated models, from 1225 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentBattler | 56.21 | Pooled Score (%) | 100 |
| CESBench - Scenario | 88.4 | Scenario score (%; the 63 open-ended engineering diagnosis i | 100 |
| Coarena - Task Completion | 90.3 | Completion Rate (%) | 100 |
| FinRiskAtlas - Evidence-Grounded Processing - Information Extraction | 85.97 | Field-Level Score (%) | 100 |
| FireWorldBench (Structured Records) | 55.34 | Completion accuracy (%; equal-weight mean of the five physic | 100 |
| FireWorldBench - Causal Mechanism Attribution (Structured Records) | 52.14 | Completion accuracy (%; P4, causal mechanism attribution ite | 100 |
| FireWorldBench - Counterfactual Intervention Reasoning (Structured Records) | 48.07 | Completion accuracy (%; P5, counterfactual intervention reas | 100 |
| FireWorldBench - Cross-Field Coupling Understanding (Structured Records) | 55.49 | Completion accuracy (%; P3, cross-field coupling understandi | 100 |
| FireWorldBench - Physical Field Perception and Grounding (Rendered Images) | 41.11 | Completion accuracy (%; P2, physical field perception and gr | 100 |
| FireWorldBench - Physical Field Perception and Grounding (Structured Records) | 54.44 | Completion accuracy (%; P2, physical field perception and gr | 100 |
| FireWorldBench - Real-World-Aligned (Rendered Images) | 51.51 | Completion accuracy (%; equal-weight mean of P1-P5 on the 71 | 100 |
| FireWorldBench - Real-World-Aligned (Structured Records) | 60.16 | Completion accuracy (%; equal-weight mean of P1-P5 on the 71 | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-5-6-luna · How It Works · Data refreshed daily, snapshot 2026-10-20.