OpenAI · 2026-03-05 · seismic
GPT-5.4 — OpenAI's frontier model with native computer use
OpenAI's most capable model: 1M-token context, native computer-use, 75% on OSWorld (first model to beat humans at desktop automation). Ships in Standard, Thinking, and Pro variants.
OpenAI's frontier model with native desktop automation that beats human experts on OSWorld — the first model to do so — plus a million-token context window.
Key specs
| Context window | 1M tokens |
|---|---|
| Price | $2.50/M input |
| Osworld verified | 75% (human: 72.4%) |
| Swe bench pro | 57.7% |
What is it?
GPT-5.4 is OpenAI's flagship model, released March 5, 2026. It ships in three variants: Standard (the everyday workhorse), Thinking (extended reasoning), and Pro (maximum capability). It is the first general-purpose model with native computer-use capabilities baked in, supporting up to 1M tokens of context in the API and Codex.
How does it work?
GPT-5.4 combines the coding capabilities of GPT-5.3-Codex with improved reasoning and professional-task handling. On OSWorld-Verified, it scores 75%, surpassing the human expert baseline of 72.4% — the first AI model to exceed human performance on desktop automation tasks. Factual accuracy improves significantly: individual claims are 33% less likely to be false and full responses are 18% less likely to contain errors compared to GPT-5.2.
Why does it matter?
Beating humans on desktop automation is a threshold moment for agentic AI. It means a model can reliably navigate software, fill forms, and complete multi-step workflows at human-or-better accuracy. Combined with the 1M context window, GPT-5.4 makes genuinely autonomous software agents practical for the first time at OpenAI's scale.
Who is it for?
Developers building agentic products, enterprise teams automating workflows, anyone using the OpenAI API.
Try it
platform.openai.com — available on all paid plans