OpenAI Launches GPT-5.4 with 1M-Token Context Window and Native Computer Use

OpenAI on March 5, 2026 released GPT-5.4, calling it their most capable and efficient frontier model for professional work. Available in ChatGPT (as GPT-5.4 Thinking), the API, and Codex, the model introduces a 1.1-million-token context window — the largest OpenAI has ever offered — and native computer-use capabilities that enable AI agents to operate software autonomously.

What Happened

GPT-5.4 arrives in three variants: the base GPT-5.4 model, GPT-5.4 Thinking (a reasoning-optimized version available in ChatGPT), and GPT-5.4 Pro for maximum performance on complex tasks. The model unifies OpenAI's recent advances in reasoning, coding, and agentic workflows into a single architecture, incorporating the coding capabilities of GPT-5.3-Codex while improving how the model works across tools, software environments, and professional tasks involving spreadsheets, presentations, and documents.

The API supports context windows up to 1.1 million tokens — roughly 750,000 words — enabling developers to feed entire codebases, legal discovery packages, or years of financial reports into a single conversation. API pricing starts at $2.50 per million input tokens and $15.00 per million output tokens, with a 10% uplift for regional data residency endpoints.

Key Details

• 1.1M-token context window — the largest OpenAI has shipped, enabling full-codebase and multi-document analysis in a single prompt
• Native computer use — GPT-5.4 is OpenAI's first general-purpose model with built-in ability to operate computers, navigate UIs, and execute multi-step workflows across applications
• Record benchmark scores — 83% on OpenAI's GDPval test for knowledge work, record scores on OSWorld-Verified and WebArena Verified computer-use benchmarks, and 75% on OSWorld-V for desktop productivity tasks
• Improved token efficiency — OpenAI reports the model solves the same problems with significantly fewer tokens than its predecessor GPT-5.2
• Three variants — base GPT-5.4, GPT-5.4 Thinking (reasoning), and GPT-5.4 Pro (maximum performance)

What Developers and Users Are Saying

Developer reaction has been mixed but generally positive. On Hacker News, the conversation centered less on benchmark supremacy and more on practical task-level performance versus competitors. Several developers noted GPT-5.4 felt meaningfully better for real coding work compared to GPT-5.2, while others preferred competing models for output formatting and conversational quality.

On Reddit, first-impression threads show a split: some users praise the speed and capability improvements, while others report an oversmart quality that makes the model harder to steer in everyday use. A recurring theme is that GPT-5.4 excels at complex professional tasks but can feel heavy-handed for simple requests.

The broader competitive context matters: app download rankings suggest growing adoption of Anthropic's Claude, and some users have publicly canceled ChatGPT subscriptions over OpenAI's Department of Defense partnership — though GPT-5.4's technical capabilities have won back some skeptics.

What This Means for Developers

The 1.1M-token context window is the headline feature for developers. This enables use cases that were previously impractical: feeding an entire monorepo into a single prompt for code review, processing complete legal or financial document sets, or building agents that maintain context across very long workflows. The native computer-use capabilities open the door for production-grade AI agents that can interact with real software — booking systems, admin panels, development environments — without custom tool integrations.

For teams already on the OpenAI API, migration is straightforward. The improved token efficiency means existing applications may see cost reductions even at the new pricing tier. Teams building agentic applications should evaluate the computer-use capabilities against their current tool-use implementations.

What's Next

OpenAI has indicated that GPT-5.4's computer-use capabilities will continue to improve, with the OSWorld benchmarks serving as the primary measure of progress. The model is available immediately in the API, ChatGPT Plus, and Codex. Developers can access the 1.1M context window through the API with no waitlist.

Sources

• OpenAI Official Blog — primary announcement and technical details
• TechCrunch — launch coverage with Pro and Thinking variant details
• The AI Insider — benchmark analysis and performance breakdown
• AI Critique — developer reactions and real-world usage analysis
• TLDL — API pricing details and token cost breakdown
• Hacker News discussion — developer community reactions