OpenAI rolls out GPT-5.4 Thinking and GPT-5.4 Pro across ChatGPT, API, and Codex

What OpenAI announced

On March 5, 2026, OpenAI said GPT-5.4 Thinking was rolling out in ChatGPT, while GPT-5.4 was also becoming available in the API and Codex. GPT-5.4 Pro is launching in ChatGPT and the API for users who want more headroom on complex work. The X post presented this as more than a routine model refresh: OpenAI is positioning GPT-5.4 as a single frontier model for reasoning, coding, and agentic workflows instead of splitting those capabilities across multiple disconnected products.

What the release page adds

The product page says GPT-5.4 is designed for professional work. OpenAI argues that the model carries forward the coding gains of GPT-5.3-Codex while improving how the system works across tools, software environments, spreadsheets, presentations, and documents. In ChatGPT, GPT-5.4 Thinking can expose an upfront plan so users can redirect the model before the final answer lands. OpenAI also highlights stronger deep web research and better context retention on tasks that require longer reasoning.

The API and Codex side is arguably the bigger operational shift. OpenAI describes GPT-5.4 as its first general-purpose model with native computer-use capabilities, which means agents can operate applications directly instead of relying only on text output plus external wrappers. The company also says GPT-5.4 supports up to 1 million tokens of context and adds tool search so agents can locate and invoke the right connectors more efficiently.

Why this matters

The practical implication is that OpenAI is compressing several previously separate value propositions into one deployment target: long-horizon reasoning, high-end coding, and direct software interaction. That makes GPT-5.4 less of a benchmark story and more of an execution story for teams building assistants and enterprise agents. If the token-efficiency gains versus GPT-5.2 hold in production, the model could also lower the cost of running more complex multi-step workflows at scale.

Sources: OpenAI X post, OpenAI