OpenAI released GPT-5.4 on March 5, its most capable system to date for professional work — and the first general-purpose model in its lineup with native, state-of-the-art computer-use capabilities, letting agents operate software and carry out workflows across applications.
One model instead of three
GPT-5.4 consolidates what had been spread across separate models: the coding strength of GPT-5.3-Codex, improved mainline reasoning, and agentic computer use, in a single model rolling out across ChatGPT, the API, and Codex. It supports up to one million tokens of context. For developers, the simplification matters as much as the capability — model selection had become its own engineering problem.
The competitive frame
Coverage of the launch consistently framed it as an assault on enterprise territory where Anthropic had built a stronghold, particularly agentic coding and professional document work. GPT-5.4's spreadsheet, presentation, and document capabilities target exactly the workflows Microsoft is simultaneously wiring into Copilot — and the workflows Anthropic's Claude Code owns among developers.
What followed
The spring cadence did not slow: GPT-5.5 arrived within weeks as OpenAI's smartest model yet, followed by GPT-5.5 Instant as ChatGPT's default, with reduced hallucinations and improved personalization. The mainline-model release cycle across the industry now runs in weeks. Enterprises building on these APIs should assume the model underneath them changes quarterly — and architect for it.