← Back
OpenAI
OpenAI releases GPT-5.4 with computer-use capabilities; achieves 83% win rate on professional knowledge work
OpenAI APIChatGPTOpenAI · releasefeaturemodelapibreaking-changeperformance · openai.com ↗

Overview

OpenAI has released GPT-5.4, the company's most capable frontier model for professional work, now available in ChatGPT (as GPT-5.4 Thinking and GPT-5.4 Pro), the OpenAI API, and Codex. The model combines advances in reasoning, coding, and agentic workflows with native computer-use capabilities—a significant milestone enabling AI agents to operate computers and execute complex workflows across applications.

Key Capabilities and Performance

Knowledge Work & Professional Tasks

  • Achieves 83.0% win rate on GDPval, matching or exceeding human professionals across 44 occupations
  • 87.3% mean score on spreadsheet modeling (vs. 68.4% for GPT-5.2)
  • Human raters prefer GPT-5.4 presentations 68% of the time due to superior aesthetics and visual design
  • 33% fewer factual errors and 18% less likely to contain any mistakes compared to GPT-5.2

Computer Use & Vision

  • Native computer-use capabilities with state-of-the-art 75.0% success rate on OSWorld-Verified (exceeding human performance at 72.4%)
  • Excellent at writing code for browser/desktop automation via Playwright and similar libraries
  • Supports 1M token context for agents to plan, execute, and verify tasks across long horizons
  • Steerable behavior via developer messages with configurable safety policies

Coding & Efficiency

  • Inherits industry-leading coding capabilities from GPT-5.3-Codex
  • 54.6% on Toolathlon benchmark (new tool-use evaluation)
  • Most token-efficient reasoning model yet, delivering faster speeds and reduced costs compared to GPT-5.2
  • Improved tool search functionality helping agents find and use the right tools more efficiently

Developer Impact & Action Items

ChatGPT Users: GPT-5.4 Thinking provides upfront thinking plans, allowing users to adjust course mid-response for more aligned outputs without additional turns. This feature is particularly useful for deep web research and maintaining context across longer conversations.

API & Codex Developers: GPT-5.4 is the first general-purpose model with native computer-use capabilities, enabling reliable agents that operate across websites and software systems. Developers can configure safety behavior via custom confirmation policies to match their risk tolerance.

Enterprise Customers: OpenAI has released a new ChatGPT for Excel add-in for enhanced spreadsheet work, alongside updated spreadsheet and presentation skills available in Codex and the API.

A higher-tier GPT-5.4 Pro variant is also available for users requiring maximum performance on complex tasks.