← Back
OpenAI
OpenAI releases GPT-5.4 with native computer-use capabilities and 83% professional knowledge work performance
OpenAI APIChatGPTOpenAI · releasefeatureapimodelperformance · openai.com ↗

Overview

OpenAI has released GPT-5.4, a major new frontier model designed for professional work across ChatGPT, the API, and Codex. This release introduces native computer-use capabilities—a first for a general-purpose OpenAI model—enabling agents to operate computers and navigate software environments autonomously. The model is available in two tiers: GPT-5.4 Thinking in ChatGPT for interactive reasoning workflows, and GPT-5.4 Pro for maximum performance on complex tasks.

Key Capabilities

Computer Use & Automation: GPT-5.4 is the first general-purpose OpenAI model with native computer-use capabilities, allowing agents to:

  • Control computers via mouse and keyboard commands in response to screenshots
  • Write code using libraries like Playwright for browser automation
  • Execute complex workflows across applications with up to 1M tokens of context
  • Achieve 75.0% success on OSWorld-Verified benchmarks (desktop navigation), surpassing human performance at 72.4% and vastly exceeding GPT-5.2's 47.3%

Professional Knowledge Work: GPT-5.4 demonstrates significant improvements on real-world professional tasks:

  • Achieves 83.0% on GDPval (professional knowledge work across 44 occupations), matching or exceeding industry professionals
  • Scores 87.3% on spreadsheet modeling tasks (vs. 68.4% for GPT-5.2)
  • Human raters preferred GPT-5.4-generated presentations 68.0% of the time over GPT-5.2 for aesthetics and visual effectiveness
  • Performs at 91% on BigLaw Bench for legal document analysis

Improved Reasoning & Accuracy:

  • GPT-5.4 Thinking in ChatGPT now displays upfront reasoning plans, allowing users to adjust course mid-response without additional turns
  • Individual claims are 33% less likely to be false compared to GPT-5.2
  • Full responses contain 18% fewer errors than GPT-5.2
  • Achieves 57.7% on SWE-Bench Pro (software engineering), up from 55.6%

Efficiency & Tools:

  • Most token-efficient reasoning model yet, using significantly fewer tokens while solving problems faster than GPT-5.2
  • Introduces tool search capability, helping agents find and use the right tools more efficiently
  • Supports 1M token context for long-horizon task planning and verification

Developer Action Items

  • API Users: GPT-5.4 is available for API consumption with native computer-use capabilities for building autonomous agents
  • ChatGPT Users: Access GPT-5.4 Thinking and Pro tiers for interactive professional work
  • Enterprise Users: New ChatGPT for Excel add-in launched today for spreadsheet work
  • Codex Users: Updated spreadsheet and presentation skills now available via GitHub