GPT-4.5: benchmark results
OpenAI's large GPT-4.5 non-reasoning model (February 2025), later retired from the API. Provider: OpenAI. Released 2025-02-27. Access: API.
Aggregate Rating 1668 ± 9, rank #274 of 1656 rated models, from 734 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM VHELM - Blink Knowledge - Forensic Detection | 97.73 | Prefix Quasi-Exact Match (%) | 100 |
| HELM VHELM - Fairness | 95.21 | Mean win rate | 100 |
| HELM VHELM - Mmmu - Psychology | 82.76 | Prefix Quasi-Exact Match (%) | 100 |
| HELM VHELM - Visual Perception | 78.37 | Mean win rate | 100 |
| ProLLM - SQL Disambiguation | 53.1 | Score (%) | 100 |
| OpenVLM MMMU - Management | 83.3 | Accuracy (%) | 99.6 |
| OpenVLM MMBench V1.1 CN - Nature Relation | 96.7 | Accuracy (%) | 99.5 |
| OpenVLM MMMU - Clinical Medicine | 80 | Accuracy (%) | 99.5 |
| OpenVLM MM-Vet - Math | 95.4 | Score | 99.3 |
| OpenVLM MMBench V1.1 CN - Spatial Relationship | 90.7 | Accuracy (%) | 99.3 |
| OpenVLM MMMU - Art and Design | 77.5 | Accuracy (%) | 99.1 |
| OpenVLM MMBench V1.1 EN - Spatial Relationship | 89.3 | Accuracy (%) | 98.9 |
Interactive version: theaggregate.ai/model?slug=gpt-4-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.