GPT-5.6 Luna (Non-reasoning): benchmark results
GPT-5.6 Luna evaluated with reasoning disabled. Provider: OpenAI. Released 2026-07-09. Access: API.
Aggregate Rating 1629 ± 13, rank #609 of 2168 rated models, from 113 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LLM Milgram Obedience - Baseline Mean Breakoff Voltage | 0 | Mean last shock administered in valid baseline sessions (vol | 98.7 |
| K-Bench - Risk (Therapeutic v0) | 93.78 | Risk Score (%, Therapeutic v0 prompt) | 88.7 |
| UGI - Natural Intelligence | 45.85 | NatInt Score | 87 |
| LLM Milgram Obedience - Baseline Full-Obedience Rate | 0 | Valid baseline sessions ending at 450 V (%; lower is less ob | 86.8 |
| K-Bench - Risk | 94.05 | Risk Score (%, Default v1 prompt) | 82 |
| OntoAct Reference - Tacit-FLE medium (ReAct) | 52.8 | Repair success rate (%) | 75 |
| UGI - Writing | 40.98 | Writing Score | 74.8 |
| AA-Omniscience Index - Software Engineering (SWE) - Go | 14 | Omniscience Index | 73.5 |
| AA-Omniscience Index - Software Engineering (SWE) - Python | 15 | Omniscience Index | 73.5 |
| AA-Omniscience Index - Software Engineering (SWE) - JavaScript | 16.36 | Omniscience Index | 72 |
| AA-Omniscience Index - Software Engineering (SWE) - Kotlin | 6 | Omniscience Index | 71.8 |
| AA-Omniscience Index - Software Engineering (SWE) - TypeScript | 12.22 | Omniscience Index | 69.9 |
Interactive version: theaggregate.ai/model?slug=gpt-5-6-luna-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.