GPT-4.5 (Preview) — benchmark results

OpenAI's GPT-4.5 research preview, a large non-reasoning model (February 2025). Provider: OpenAI. Released 2025-02-27. Access: API.

Unified ELO 1658 ± 23, rank #351 of 1776 rated models, from 79 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Smolagents Leaderboard78.75Average Accuracy (%)100
AI Chess Leaderboard (Continuation)1800Elo99.6
KOFFVQA - Commonsense Reasoning93.56Score (%)98.8
Smolagents - GAIA56.25Accuracy (%)98.1
Smolagents - SimpleQA88Accuracy (%)98.1
BenchTable78.5Total Score (%)95.8
KOFFVQA - Object Attributes85.5Score (%)95.1
KOFFVQA - Korean OCR95Score (%)93.8
AidanBench3003Novel Answers92.7
KOFFVQA - Table Understanding93.33Score (%)91.4
KOFFVQA - Document Understanding90.67Score (%)88.9
PlatinumBench (MIT)0.99Avg Error Rate (%)87.9

Interactive version: theaggregate.ai/model?slug=gpt-4-5-preview · How the rankings work · Data refreshed daily, snapshot 2026-07-22.