GPT-5.4 brings computer use into a general purpose model

OpenAI released GPT-5.4 on 5 March in ChatGPT, the API and Codex, with a Pro version for more demanding work. Its launch post described this as its first general purpose model with native computer-use capabilities in the API and Codex.

The model incorporated coding abilities from GPT-5.3-Codex and added tool search, allowing an agent to find relevant tools in a larger collection. In ChatGPT, the Thinking version could show an initial plan so users could redirect a response while it was underway.

OpenAI offered context of up to one million tokens, with experimental configuration required for the larger window in Codex. Standard API rates were $2.50 per million input tokens and $15 per million output tokens. The company’s claim of greater token efficiency does not make every workflow cheaper; the actual bill depends on the tokens, tools and context a task consumes. Longer inputs and outputs still need to be budgeted against the stated rates.