OpenAI on Thursday introduced GPT-5.4, a new flagship artificial intelligence model rolling out across ChatGPT, the company’s API and Codex as it pushes further into workplace software, coding and software agents. In ChatGPT, the release appears as GPT-5.4 Thinking, while a higher-performance GPT-5.4 Pro version is being offered for more demanding tasks.
The company said GPT-5.4 is aimed at professional work and combines improvements in reasoning, coding and multi-step automated workflows. OpenAI said the model is better suited for tasks involving spreadsheets, presentations and documents, and that in ChatGPT it can now present an upfront plan for longer requests so users can redirect the response while it is still being generated. The company also said the model improves web research for highly specific queries and keeps context better during longer tasks.
OpenAI positioned the release as a step up from GPT-5.2 on several benchmarks tied to office work, web research and computer use. It said GPT-5.4 scored 83.0% on GDPval, compared with 70.9% for GPT-5.2, and 75.0% on OSWorld-Verified, compared with 47.3% for GPT-5.2. The company also said GPT-5.4’s individual claims were 33% less likely to be false and its full responses were 18% less likely to contain any errors on prompts where users had previously flagged factual mistakes.
For developers, OpenAI said GPT-5.4 is its first general-purpose model with native computer-use capabilities, allowing agents to navigate websites and software systems, interpret screenshots and carry out tasks across applications. The company said the model supports up to 1 million tokens of context and adds a feature called tool search, designed to help agents work more efficiently across large tool libraries. In one internal demonstration using 250 benchmark tasks, OpenAI said tool search reduced total token usage by 47% while maintaining the same accuracy.
OpenAI also tied the release to its safety framework. The company said it is treating GPT-5.4 as having “High cyber capability,” the same classification it used for GPT-5.3-Codex, and is deploying the model with monitoring systems, trusted access controls and blocking measures for some higher-risk requests on certain Zero Data Retention surfaces. OpenAI said it is also continuing work on methods for monitoring reasoning and detecting potential misuse.
GPT-5.4 began rolling out Thursday in ChatGPT and Codex and is available immediately in the API as gpt-5.4, with gpt-5.4-pro offered for developers seeking higher performance on complex tasks. In ChatGPT, GPT-5.4 Thinking is available to Plus, Team and Pro users, replacing GPT-5.2 Thinking. OpenAI said GPT-5.2 Thinking will remain available for paid users as a legacy option until June 5, 2026, while Enterprise and Edu customers can enable early access through admin settings. GPT-5.4 Pro is available on Pro and Enterprise plans.
OpenAI said the new model is priced above GPT-5.2 in the API, though it argued that greater token efficiency could reduce total usage on many tasks. The company listed GPT-5.4 at $2.50 per million input tokens and $15 per million output tokens, compared with $1.75 and $14 for GPT-5.2. GPT-5.4 Pro is priced at $30 per million input tokens and $180 per million output tokens.
OpenAI said it chose the GPT-5.4 name because the model is its first mainline reasoning release to incorporate the coding capabilities introduced in GPT-5.3-Codex, and because the company wants to simplify model choices for users working in Codex.
