GPT-4.5 Preview

OpenAI · February 2025

deprecatedClosedmixture of expertsmultimodalAPI Available
Context Window128K tokens
Sunset DateJune 2026

Description

A research preview that explored massive unsupervised learning rather than reasoning improvements. Focused on emotional intelligence, creativity, and natural-sounding open-ended writing — areas where earlier models felt mechanical. Not a reasoning model; designed to feel more human in conversation.

Benchmark Scores

MMLUMassive Multitask Language Understanding — 57 subjects
85.1%
GPQAGraduate-level science QA
71.4%
SWE-benchReal-world software engineering
38.0%

Key Innovations

Multimodal
MultimodalProcessing multiple types of input (text, images, audio, video) in a single model.

Family Tree

Built On

External Links