Major Release OpenAI Multimodal Proprietary

OpenAI GPT-4V (Vision)

GPT-4 gains eyes — the model that made multimodal mainstream

Released 2023-09-25 · ~175B · 128K tokens context · knowledge cutoff 2023-04

Overview

GPT-4 enhanced with vision capabilities for image understanding

Specifications

DeveloperOpenAI
Release date2023-09-25
Model typeMultimodal
Parameters~175B
ArchitectureTransformer + vision encoder
Context window128K tokens
Knowledge cutoff2023-04
LicenseProprietary
Input modalitiestext, image
Output modalitiestext

Benchmarks

Benchmark GPT-4V (Vision)
MMLU 86.4%
MathVista 49.9%
AI2D 78.2%

Availability

API Only OpenAI API Azure OpenAI

Official references