GPT-4o Mini: benchmark results
OpenAI's small, low-cost multimodal GPT-4o tier with a 128K context that replaced GPT-3.5 Turbo in ChatGPT (July 2024). Provider: OpenAI. Released 2024-07-18. Access: API.
Aggregate Rating 1555 ± 3, rank #599 of 1656 rated models, from 2102 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ActiveSciBench-Chem (LLM-AutoSciLab) | 35.09 | Symbolic accuracy (%; share of the 57 enzyme-kinetics tasks | 100 |
| ActiveSciBench-GRN (LLM-AutoSciLab) | 31.11 | Exact graph accuracy (%; share of the 45 gene-regulatory-net | 100 |
| ActiveSciBench-GRN (LLM-AutoSciLab) - Edge F1 | 72.49 | Edge F1 (%; directed-edge recovery against the hidden regula | 100 |
| BoxComm (Punch Events) - Play-by-Play | 18.5 | Content consistency accuracy (%) on the sentences whose requ | 100 |
| BoxComm - Play-by-Play | 15.2 | Content consistency accuracy (%) on the sentences whose requ | 100 |
| CRMArena - HTU (Act) | 38.5 | Exact match (%) | 100 |
| ClinTraceBench - Retrospective Propagation | 28.3 | Accuracy (%; task T4, retrospective propagation: list the ea | 100 |
| Czech Opinion ABSA (Zero-Shot) | 44.1 | Mean of the ASTE, ASQP and ACOS micro F1 scores (%) on the C | 100 |
| Czech Opinion ABSA (Zero-Shot) - ASQP | 42.12 | Micro F1 (%) of exact-match aspect sentiment quad (aspect te | 100 |
| Czech Opinion ABSA (Zero-Shot) - ASTE | 56.43 | Micro F1 (%) of exact-match aspect sentiment triplet (aspect | 100 |
| DBLifeBench - Progressive Text2SQL | 54.31 | Graph execution accuracy (%; mean node execution accuracy ov | 100 |
| FDARxBench (Oracle Passages) - Multi-Hop | 82.2 | Answer accuracy (x 100) on multi-hop questions from FDARxBen | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-4o-mini · How It Works · Data refreshed daily, snapshot 2026-10-20.