OpenAI GPT-5.4: Computer Use and 1M Tokens
OpenAI releases GPT-5.4 in standard, Thinking, and Pro variants — its most capable model yet, with native computer-use mode, 1 million token context, 33% fewer errors, and direct Excel and Google Sheets integration.
OpenAI has launched GPT-5.4, calling it "our most capable and efficient frontier model for professional work." Released on March 5, 2026, the model is available in three variants — standard, Thinking, and Pro — and introduces native computer-use capabilities, a 1 million token context window, and direct integration with Microsoft Excel and Google Sheets. It's the top story on Techmeme today.
GPT-5.4 Thinking and GPT-5.4 Pro are rolling out now in ChatGPT.
— OpenAI (@OpenAI) March 5, 2026
GPT-5.4 is also now available in the API and Codex.
GPT-5.4 brings our advances in reasoning, coding, and agentic workflows into one frontier model.
Three Tiers, One Model
GPT-5.4 comes in three tiers. GPT-5.4 (standard) is available to all ChatGPT users. GPT-5.4 Thinking adds deep reasoning with interruptible chains — available to Plus, Team, and Enterprise subscribers. GPT-5.4 Pro is the high-performance variant, reserved for ChatGPT Pro ($200/month) and Enterprise plans.
The headline features:
Computer use. GPT-5.4 is the first OpenAI model with native computer-use mode. Available in Codex and through the API, it can interact with screens, manipulate files, and control applications. This earned record scores on computer-use benchmarks OSWorld-Verified and WebArena Verified.
1 million token context. The API version supports up to 1M tokens of input — the largest context window OpenAI has ever offered. Pricing doubles past 272,000 tokens.
Improved accuracy. OpenAI claims 33% fewer individual claim errors and 18% fewer full-response errors compared to GPT-5.2. The model scored 83% on GDPval, OpenAI's internal benchmark for knowledge work tasks.
Office integration. GPT-5.4 introduces direct plugins for Microsoft Excel and Google Sheets, enabling financial analysis, data manipulation, and formula generation from within ChatGPT.
How It Works
The standard model handles everyday tasks — writing, analysis, search, coding. Thinking mode activates extended reasoning chains for complex problems; a new feature lets you interrupt the model mid-thought and redirect its direction, which is useful when you can see it heading in the wrong direction. Pro mode adds maximum computational resources for the hardest problems.
Computer use works through a vision-action loop: the model sees the screen, decides what to do, executes actions (clicks, typing, file operations), and observes the result. This runs in Codex's sandboxed environment or through the API for custom deployments.
Token efficiency is a quiet but important upgrade. OpenAI says GPT-5.4 solves the same problems with significantly fewer tokens than GPT-5.2, which partially offsets the higher per-token pricing.
What It Costs
API pricing reflects the capability increase. GPT-5.4 standard: $2.50 per million input tokens, $15 per million output tokens. GPT-5.4 Pro: $30 per million input tokens, $180 per million output tokens. Input tokens beyond 272K are charged at double rates.
For ChatGPT users, GPT-5.4 Thinking is included with Plus ($20/month) and Team subscriptions. Pro requires the $200/month ChatGPT Pro plan.
| Spec | GPT-5.4 | GPT-5.4 Thinking | GPT-5.4 Pro |
|---|---|---|---|
| Context | 1M tokens | 1M tokens | 1M tokens |
| Computer use | API/Codex | API/Codex | API/Codex |
| API input | $2.50/M | — | $30/M |
| API output | $15/M | — | $180/M |
| ChatGPT tier | All users | Plus/Team/Enterprise | Pro ($200/mo) |
The Unification Play
GPT-5.4 is OpenAI unifying its product line. Previous generations had separate model families for reasoning (o-series) and general use (GPT series) — think of how GPT-5.3 Instant offered a lighter, faster variant while maintaining capability gaps between models. GPT-5.4 merges both into one model with a thinking dial, simplifying the choice for developers and users.
Computer use is the strategic play. By shipping it as a native capability in their flagship model, OpenAI is positioning GPT-5.4 as the foundation for autonomous agents that don't just answer questions but actually do work — fill out forms, navigate apps, manage files, execute multi-step workflows. This mirrors the vision behind projects like Symphony, which orchestrates coding agents at the project management level.
The Excel and Sheets integration signals where OpenAI thinks the money is: enterprise knowledge workers. If GPT-5.4 can reliably analyze spreadsheets, generate reports, and handle financial modeling directly in the tools people already use, that's a much more compelling enterprise pitch than a standalone chat interface. This approach contrasts with how Google is positioning Canvas in AI Mode — embedding creation directly into search rather than standalone productivity apps.
Full details on OpenAI's blog.