GPT-5 Mini: benchmark results
OpenAI's smaller, cost-efficient GPT-5 tier. Provider: OpenAI. Released 2025-08-07. Access: API.
Aggregate Rating 1704 ± 3, rank #186 of 1656 rated models, from 2044 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AJ-Bench (Agent-as-a-Judge) - FileSystem | 75.8 | F1 (%) on FileSystem data-system tasks (MCPMark; 24 tasks, 1 | 100 |
| AJ-Bench (Agent-as-a-Judge) - Wide Search | 74.48 | F1 (%) on Wide search (WideSearch-derived item checks; 9 tas | 100 |
| AMA-Bench - Open-World QA | 78.81 | Average Score (%) | 100 |
| AMA-Bench - Text2SQL | 83.44 | Average Score (%) | 100 |
| AMA-Bench - Web | 82.95 | Average Score (%) | 100 |
| AgentEHR (ReAct) - Diagnoses | 42.7 | F1 (%; predicted diagnoses label set against the next record | 100 |
| AgentEHR (ReAct) - Labevents | 21.51 | F1 (%; predicted labevents label set against the next record | 100 |
| AgentSocialBench - Cross-Domain Leakage | 0.4 | Privacy leakage rate (0-1, share of designated private items | 100 |
| AgentWebBench (Embedding-Selected Tools) - Deep Research | 70.87 | Key Point Recall (%) of long reports on 331 DeepResearchGym | 100 |
| AgentWebBench (LLM-Selected Tools) - Deep Research | 70.62 | Key Point Recall (%) of long reports on 331 DeepResearchGym | 100 |
| AgentWebBench (Multi-Agent Coordination) - Deep Research | 69.39 | Key Point Recall (%) of long reports on 331 DeepResearchGym | 100 |
| AsyncPlan Online Robo Challenge (Planner) | 35 | Plan accuracy (%; share of the 140 Online Robo Challenge epi | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-5-mini · How It Works · Data refreshed daily, snapshot 2026-10-20.