2026-03-07

gpt-5-4 - OpenAI's Frontier Model with Native Computer Use

OpenAI's most capable frontier model — first general-purpose model with native computer-use capabilities (75.0% OSWorld, exceeds human 72.4%), 1M context window, tool search (47% token reduction), combines GPT-5.3-Codex coding with advanced reasoning; 83.0% on GDPval (professional work across 44 occupations vs 70.9%), upfront planning with mid-response steering in ChatGPT, experimental Playwright Interactive skill for visual debugging. Available in ChatGPT (Thinking mode), API ($2.50/$15 per M tokens), and Codex with fast mode (1.5x token velocity).