OpenAI Rolls Out GPT-5.4 Across ChatGPT, the API, and Codex with 1M Context and Native Computer Use

X rollout and core positioning

On March 5, 2026, OpenAI posted that GPT-5.4 Thinking and GPT-5.4 Pro are rolling out across ChatGPT, the API, and Codex. That is not just a model refresh for one surface. It is a coordinated release across consumer chat, developer APIs, and agentic coding workflows, which usually signals that OpenAI views the model as a general platform foundation rather than a niche specialty variant.

What OpenAI says is new in GPT-5.4

OpenAI’s launch article describes GPT-5.4 as its most capable and efficient frontier model for professional work. The company says it combines the coding strengths of GPT-5.3-Codex with improvements in reasoning, tool use, computer interaction, and knowledge-work tasks involving spreadsheets, presentations, and documents. One of the biggest practical changes is scale: GPT-5.4 supports up to 1M tokens of context, which OpenAI says helps agents plan, execute, and verify tasks over longer horizons.

The company also calls GPT-5.4 its first general-purpose model with native computer-use capabilities. In the launch article, OpenAI says the model can operate computers, interact with websites and software environments, and work with larger tool ecosystems through tool search, which lets the model look up tool definitions when needed instead of loading everything into context upfront. That matters for agent builders because large MCP and tool ecosystems can otherwise inflate prompts, raise cost, and slow down responses.

Benchmarks and workflow implications

OpenAI cites several benchmark gains, including 83.0% on GDPval, 57.7% on SWE-Bench Pro, 75.0% on OSWorld-Verified, and 82.7% on BrowseComp. These are vendor-reported numbers, but the mix is notable: they cover knowledge work, coding, browser or desktop interaction, and persistent web search rather than only one benchmark category. OpenAI is clearly framing GPT-5.4 as an agent model for real-world professional workflows, not just a stronger chatbot.

The article also emphasizes spreadsheet, document, and presentation work. OpenAI says GPT-5.4 improves factual accuracy relative to GPT-5.2 and shows stronger performance on internal spreadsheet-modeling and presentation evaluations. In practice, that suggests the release is aimed at enterprise workflows where agents have to combine reasoning, tool use, and formatted deliverables rather than only answer questions in plain text.

Why this matters now

The high-signal part of this launch is the packaging of several capabilities that often arrive separately: long context, computer use, tool search, coding quality, and professional document work. By placing the same model across ChatGPT, the API, and Codex, OpenAI is reducing the gap between experimentation and deployment. Teams can prototype with the same model family they expect to use in production agents and coding systems, while end users get a parallel upgrade in the chat surface.

Primary sources: X post, OpenAI launch article.