AI Hub
All papers
ArchitectureEvaluationSafety

GPT-4 Technical Report

OpenAI·March 15, 2023

OpenAI

View on arXiv

TL;DR

Reports GPT-4, a large multimodal model with human-level performance on many professional and academic benchmarks, along with its evaluation and safety work — but, notably, not its architecture or training details.

Why it matters

Marked the moment frontier LLMs became broadly capable professional tools — and the moment leading labs stopped publishing technical details, citing competition and safety.

Related models

Related terms