GPT-5.4: Native Computer Use, 1M Context Window and Tool Search

GPT-5.4 Arrives in Codex

OpenAI released GPT-5.4 on March 5, 2026, bringing its most capable frontier model to the full Codex surface area — the desktop app, CLI, IDE extension, Codex Cloud, and the API.

Native Computer Use

GPT-5.4 is the first general-purpose OpenAI model with native, state-of-the-art computer-use capabilities. The model ships with it built in, allowing agents to interact directly with desktops and applications, carry out multi-step workflows, and complete a build-run-verify-fix loop without human hand-holding.

1M-Token Context Window

GPT-5.4 supports an experimental context window of up to 1 million tokens — doubling the previous ceiling and making it practical to analyze entire codebases in a single request.

Improved Reasoning and Efficiency

GPT-5.4 is available in three variants: standard, Thinking, and Pro. Across coding benchmarks, GPT-5.4 matches or outperforms GPT-5.3-Codex on SWE-Bench Pro while using approximately 47% fewer tokens. The model is also 33% less likely to make factual errors compared to GPT-5.2.

Tool Search

GPT-5.4 adds a tool search capability that helps agents find and invoke the correct connector or skill within large ecosystems of MCP servers, plugins, and app connectors.

Availability

GPT-5.4 is available immediately across all Codex surfaces. Use codex --model gpt-5.4 in the CLI, or select from the model picker in the IDE extension or app.