GPT-5.3: benchmark results
OpenAI's GPT-5.3 Instant (March 2026), a fast everyday-conversation chat model in the GPT-5 series focused on smoother, more natural responses for common ChatGPT use. Provider: OpenAI. Released 2026-03-03. Access: API.
Aggregate Rating 1760 ± 10, rank #103 of 1656 rated models, from 287 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HelpBench - Data Concerns | 89 | Data Concerns topic (data collection, exposure, loss and sha | 100 |
| HelpBench - Moderation Tools | 84 | Moderation Tools topic (moderation and anti-abuse tools such | 100 |
| MapSatisfyBench - Explicit Completion | 92.72 | Explicit-factor completion rate (%): share of the query's st | 100 |
| PerfReasoning - Default Effort | 3.2 | Both-Exact Pass Rate (%) | 100 |
| PhysAI-Bench | 52 | Decision accuracy (%; four-option choice of the next action | 100 |
| Synthetic Hospital - Evidence Retrieval | 83.3 | Precision at 5 (%; ranking chart sections for a diagnosis ag | 100 |
| Synthetic Hospital - Imaging Indication | 51.8 | Concept F1 (%; inferring the clinical question behind a vagu | 100 |
| Evals for Every Language - Basque (eu) | 74.45 | Average Score (%) | 98.4 |
| Evals for Every Language - Southern Sotho (st) | 72.24 | Average Score (%) | 98.4 |
| Evals for Every Language - Tswana (tn) | 67.41 | Average Score (%) | 98.4 |
| HelpBench | 87 | Overall rubric score (%): points for met positive criteria a | 97.1 |
| HelpBench - Compromise | 90 | Compromise topic (compromised accounts and devices, malware | 97.1 |
Interactive version: theaggregate.ai/model?slug=gpt-5-3 · How It Works · Data refreshed daily, snapshot 2026-10-20.