GPT-5.3 Instant: benchmark results
Provider: OpenAI. Released 2026-03-03. Access: API.
Aggregate Rating 1827 ± 30, rank #47 of 1635 rated models, from 30 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MedHarm | 0.2 | Unsafe response rate (%): share of queries whose response co | 100 |
| MedHarm - Actionable Harm Rate | 0.1 | Actionable harm rate (%): share of queries answered with con | 100 |
| MedHarm - Refusal Adequacy | 99.9 | Refusal adequacy (%): mean judge score for refusals that blo | 100 |
| MedHarm - Safe Helpfulness | 99.7 | Safe helpfulness (%): mean judge score for responses that av | 100 |
| OpenAI GPT-5.6 August Update System Card - Adversarial User Simulations - Emotional Reliance | 98.7 | not_unsafe message rate (%) | 100 |
| OpenAI GPT-5.6 August Update System Card - Challenging Prompts - Extremism | 94.3 | not_unsafe rate (%) | 100 |
| OpenAI GPT-5.6 August Update System Card - Image Input - Erotic | 99.8 | not_unsafe rate (%) | 100 |
| OpenAI GPT-5.6 August Update System Card - Image Input - Extremism | 98.9 | not_unsafe rate (%) | 100 |
| OpenAI GPT-5.6 August Update System Card - Image Input - Hate | 99.8 | not_unsafe rate (%) | 100 |
| UAV-DualCog - Landmark-Driven Action Decision | 62.3 | Answer accuracy (%; 1,024 questions asking which way the UAV | 100 |
| UAV-DualCog - Landmark-Relative Direction | 56.5 | Answer accuracy (%; 1,024 questions asking where a landmark | 91.4 |
| UAV-DualCog - Future Observation Prediction | 37.3 | Answer accuracy (%; 1,024 questions asking which view the UA | 88.2 |
Interactive version: theaggregate.ai/model?slug=gpt-5-3-instant · How It Works · Data refreshed daily, snapshot 2026-11-05.