TL;DR
OpenAI has released GPT-5.4, its most capable model to date. The release combines frontier coding from GPT-5.3-Codex with native computer-use capabilities, a one-million-token context window, and a new tool search feature. GPT-5.4 is available in ChatGPT, the API, and Codex, with a Pro tier for maximum performance on complex tasks.
What GPT-5.4 Does
The headline feature is native computer use. GPT-5.4 is OpenAI’s first general-purpose model that can operate computers directly — navigating browsers, clicking UI elements, and executing multi-step workflows across applications. On OSWorld-Verified, a benchmark for desktop navigation, it scored 75.0%, exceeding human performance at 72.4%.
The model also supports up to one million tokens of context, allowing agents to plan and execute tasks across much longer horizons. A new tool search feature lets the model work with large tool ecosystems efficiently, reducing token usage by 47% on multi-tool benchmarks while maintaining accuracy.
On professional work, GPT-5.4 matched or exceeded industry professionals in 83.0% of comparisons on GDPval, which tests across 44 occupations. It scored 87.3% on investment banking modelling tasks, up from 68.4% for GPT-5.2, and hallucinations dropped by 33% at the individual claim level.
Coding and Developer Tools
GPT-5.4 matches or beats GPT-5.3-Codex on coding benchmarks while adding the computer-use and professional capabilities. A new /fast mode in Codex delivers 1.5x faster output. An experimental Playwright skill lets the model visually debug web applications, including testing apps it is actively building.
Pricing
API pricing comes in at $2.50 per million input tokens and $15 per million output tokens — higher than GPT-5.2 ($1.75/$14) but offset by the model’s greater token efficiency. GPT-5.4 Pro, for maximum performance on complex tasks, costs $30/$180 per million tokens.
Safety Notes
OpenAI is treating GPT-5.4 as “High cyber capability” under its Preparedness Framework. The company published a new evaluation for chain-of-thought controllability, finding that GPT-5.4’s ability to deliberately hide its reasoning is low — a positive sign for safety monitoring.
Looking Forward
GPT-5.4 represents OpenAI’s push toward autonomous agents that can operate computers, use tools, and complete professional work with minimal supervision. For businesses evaluating AI tools, the computer-use capabilities and million-token context may open up workflows that were previously impractical, though pricing at the Pro tier puts maximum performance at a premium.