OpenAI launched GPT-5.4 across ChatGPT, the API, and Codex on Thursday. The launch post bills it as "our most capable and efficient frontier model for professional work," which is the standard release copy. The genuinely new addition: GPT-5.4 is OpenAI's first mainline model with native computer use, letting it interact directly with software to complete, verify, and fix tasks without a separate tool layer. It also folds in the coding capabilities from GPT-5.3-Codex, which previously lived in a separate model.
In ChatGPT, GPT-5.4 Thinking is live today for Plus, Team, and Pro subscribers, replacing GPT-5.2 Thinking as the default reasoning option. GPT-5.4 Pro is restricted to Pro and Enterprise. Android access is live; iOS is still pending. Enterprise and Edu admins need to flip the early access toggle in workspace settings. Developers get both gpt-5.4 and gpt-5.4-pro in the API immediately, with a 1-million-token context window available in Codex.
On accuracy, OpenAI reports individual claims are 33% less likely to be false compared to GPT-5.2, and full responses are 18% less likely to contain any error. Both figures are company-measured. The model scored 83% on OpenAI's GDPval knowledge-work eval spanning 44 occupations, up from an earlier reported 70.9% for GPT-5.2 Thinking. GPT-5.2 Thinking stays available in the legacy model picker until June 5, 2026.
For API developers, a new Tool Search system replaces the old approach of loading all tool definitions into the context upfront, which should cut token overhead in large agent setups.
Bottom Line
GPT-5.2 Thinking is now on a three-month retirement clock, with its sunset date set for June 5, 2026.
Quick Facts
- Released: Thursday, March 5, 2026
- API model strings: gpt-5.4, gpt-5.4-pro
- Context window: 1M tokens in Codex (experimental)
- Hallucination reduction: 33% fewer false claims vs. GPT-5.2 (company-reported)
- GPT-5.2 Thinking retirement date: June 5, 2026
