OpenAI released GPT-5.4 on Thursday, March 5, 2026, introducing the first general-purpose language model with native computer-use capabilities and support for 1 million token context windows. The model enables AI agents to operate computers and execute complex workflows across applications, representing a significant advancement in autonomous task completion.
GPT-5.4 Achieves 83% Performance on Professional Work Benchmarks
GPT-5.4 delivers substantial improvements in accuracy and professional capability over its predecessor. The model is 33% less likely to make errors in individual claims compared to GPT-5.2, with overall responses 18% less likely to contain errors. On GDPval, which tests agent performance across 44 occupations, GPT-5.4 achieved 83.0% performance matching or exceeding industry professionals, up from 70.9% for GPT-5.2.
Key technical specifications include:
- Context windows up to 1 million tokens in the API version
- 75% score on computer use benchmarks, the highest recorded to date
- 47% reduction in total token usage with tool-search configuration while maintaining accuracy
- Native integration with Microsoft Excel and Google Sheets for automated analysis
Multiple Model Versions Available Across ChatGPT and API
OpenAI released GPT-5.4 in two primary configurations: GPT-5.4 Thinking and GPT-5.4 Pro. GPT-5.4 Thinking is available to ChatGPT Plus, Team, and Pro users, while GPT-5.4 Pro is limited to Pro and Enterprise plans. Both versions are also accessible through the OpenAI API and Codex platform.
The model introduces new capabilities for knowledge work, including mid-response steering and enhanced web search functionality. OpenAI also launched a suite of ChatGPT integrations that allow GPT-5.4 to connect directly to spreadsheet applications for granular analysis and automated task completion.
Key Takeaways
- GPT-5.4 is the first general-purpose model with native computer-use capabilities, scoring 75% on computer use benchmarks
- The model supports context windows up to 1 million tokens, enabling agents to plan and execute complex, long-horizon tasks
- GPT-5.4 achieves 83.0% performance on GDPval professional work benchmarks, 12.1 percentage points higher than GPT-5.2
- Error rates decreased by 33% for individual claims and 18% for overall responses compared to GPT-5.2
- The model is available now through ChatGPT (Plus, Team, Pro, and Enterprise), the OpenAI API, and Codex