OpenAI Launches GPT-5.4: A New Frontier Model Built for Native Computer Use and Professional Tasks

Today OpenAI announced the release of GPT-5.4, its most efficient frontier model designed specifically for professional work. Available across ChatGPT, the API, and Codex, this new iteration consolidates the reasoning power of GPT-5.2 with the specialized coding skills of GPT-5.3-Codex. This launch also includes a high-performance Pro variant for users tackling the most complex long-horizon tasks.

The standout feature of GPT-5.4 is its native computer-use capability, which allows the model to operate websites and software directly. By interpreting screenshots and issuing mouse or keyboard commands, it achieved a 75.0% success rate on the OSWorld-Verified benchmark. This performance significantly outpaces the 47.3% score of GPT-5.2 and even exceeds human baselines. More details on these performance leaps are available at https://openai.com/news/product-releases/.

Efficiency is a primary focus of this update, highlighted by the introduction of the new Tool Search feature. Instead of crowding the context window with every possible tool definition, the model now searches for the specific tools it needs on demand. This approach reduces token usage by nearly 47% in heavy workflows while maintaining high accuracy across multi-step tasks. Developers can also leverage a massive 1 million token context window, enabling agents to plan and execute tasks over much longer periods.

Coding workflows see a significant boost, with a new fast mode in Codex that delivers 1.5x faster token velocity. These technical improvements aim to reduce the back and forth often required with earlier, less capable models. Steerability has also been upgraded in https://chatgpt.com/, where GPT-5.4 Thinking provides an upfront plan of its logic. Users can now adjust the model direction mid-response to ensure the final output aligns perfectly with their requirements.

Availability starts today for ChatGPT Plus, Team, and Pro users, while Enterprise and Edu customers can access the model via their admin settings. GPT-5.4 is priced slightly higher per token in the API to reflect its capabilities, though OpenAI notes the increased token efficiency should lower overall costs for many users. The legacy GPT-5.2 model will remain available until June 2026 for a smooth transition.