OpenAI launches GPT‑5.4 with native computer use
SkimNews Take
Native computer control reframes the 33% error reduction: every remaining hallucination now executes as a real action across other applications, compounding consequences in ways text-only models never triggered.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI launched GPT-5.4, a model that combines reasoning, coding, and professional work with spreadsheets, documents, and presentations.
- GPT-5.4 is OpenAI's first model with native computer use capabilities, enabling it to operate a computer on the user's behalf across applications.
- GPT-5.4 Thinking is being rolled out to ChatGPT Plus, Team, and Pro users, providing an outline of its work and allowing users to tweak requests during the response.
- GPT-5.4 can write code to operate computers, issue keyboard and mouse commands from screenshots, and shows improved web browsing and tool/API usage.
- GPT-5.4 can more persistently search across multiple rounds to find relevant sources for “needle‑in‑a‑haystack” questions and synthesize clear, well‑reasoned answers.
- OpenAI says GPT-5.4 is its most factual model yet, with individual claims 33 % less likely to be false than GPT‑5.2.
- GPT-5.4 Pro is being released via the API and for ChatGPT Enterprise and Edu users, targeting maximum performance on complex tasks.
Why it matters: Developers and enterprise users gain a model that can directly control computers and browse the web, reducing the need for manual scripting and speeding up complex tasks, while OpenAI strengthens its position in the emerging autonomous‑agent market and can monetize new API services.
