OpenAI Releases GPT-5.4: Computer Use, 1 Million Token Context and Stronger Reasoning

OpenAI has released GPT-5.4, its newest and most capable model to date. The model combines the best of previous specialized versions into one package — and sets a new standard for what AI can do in practice.

What's New in GPT-5.4?

GPT-5.4 is available in two variants: GPT-5.4 Thinking for everyday use and GPT-5.4 Pro for demanding tasks. Both are available via ChatGPT and the OpenAI API.

Computer Use – AI Takes Over the Desktop

Perhaps the most groundbreaking feature is "Computer Use" — the model can operate desktop applications, click menus, fill out forms, and complete tasks autonomously. In tests, GPT-5.4 solved 75% of desktop tasks, marginally higher than the human score of 72.4%.

1 Million Token Context Window

Via API and Codex, the model supports up to one million tokens in a single session — equivalent to hundreds of research papers, full contract libraries, or large codebases.

Fewer Errors

Individual claims are 33% less likely to be incorrect compared to GPT-5.2. Overall responses are 18% less likely to contain errors.

Automatic Tool Selection

The new "Tool Search" engine lets the model identify and use the right tools without developers needing to list them in every prompt.

Excel and Google Sheets Integration

Plugins allow the model to connect directly to spreadsheets, update cells, run calculations, and complete tasks without manual copy-pasting.

Efficiency and Costs

GPT-5.4 uses up to 47% fewer tokens on certain tasks compared to previous versions, meaning lower costs for developers and businesses using AI at scale.

Benchmarks

• 83% – Matches professionals across 44 occupations
• 75% – OSWorld desktop navigation (beats humans)
• 33% – Fewer hallucinations vs GPT-5.2
• 1M – Token context window in API and Codex

GPT-5.4 marks a shift from AI that answers questions to AI that performs real tasks in the real world.