Atlas

Models

← All models

GPT-4 Turbo with Vision

OpenAI · 2023-11-06

GPT-4 Turbo with Vision was OpenAI's preview API model for text-and-image input, exposed through Chat Completions as gpt-4-vision-preview. Announced at DevDay on November 6, 2023, it supported a 128,000-token context window and text outputs for tasks such as captioning, detailed image analysis, and reading documents with figures. OpenAI shut the model down on December 6, 2024 and recommended GPT-4o as its replacement.

Benchmark scores

BenchmarkScore
Capability117.9
MATH-Vision22.8
MathVista (mini)49.9
MMMU56.8
OSWorld (self-reported, Screenshot + A11y tree)12.2
OSWorld (self-reported, Set-of-Mark)11.8
Video-MME59.9
Loading Atlas data…