OpenAI rolls out GPT‑5.4, promising 1M‑token context and native computer use

OpenAI has released GPT‑5.4, a new frontier model it says consolidates recent gains in reasoning, coding and agentic workflows and is rolling the technology across ChatGPT, its developer API and the Codex platform. The company published a product blog on March 5, 2026 describing GPT‑5.4 as bringing together "the best of our recent advances in reasoning, coding, and agentic workflows into a single frontier model."

The release introduces two branded variants: a ChatGPT variant called GPT‑5.4 Thinking and a higher performance offering, GPT‑5.4 Pro. OpenAI and reporting by technology outlets agree the model family is available across ChatGPT, Codex and the API, but accounts differ on which customer plans get which variant. Cryptobriefing reported that GPT‑5.4 Thinking is available to ChatGPT Plus, Team and Pro subscribers while GPT‑5.4 Pro is limited to Pro and Enterprise plans and to developers via the API. Gizmodo and other outlets published similar but not identical availability claims for enterprise and education subscribers.

OpenAI highlighted several technical advances. The model supports up to 1,000,000 tokens of context, which the company said allows agents "to plan, execute, and verify tasks across long horizons." For the first time in a general-purpose release, OpenAI said GPT‑5.4 has native computer‑use capabilities that let agents interact with operating systems, websites and applications via mouse, keyboard and visual inputs, enabling them to operate software and carry out complex workflows across applications.

The company also emphasized improvements in how models find and call external tools. "GPT‑5.4 also improves how models work across large ecosystems of tools and connectors with tool search, helping agents find and use the right tools more efficiently without sacrificing intelligence," OpenAI wrote. It said tool calling is more accurate and efficient, and that GPT‑5.4 "achieves higher accuracy in fewer turns on Toolathlon," a benchmark that measures multi‑step tool use. OpenAI illustrated a Toolathlon task in its announcement: "an agent needs to read emails, extract assignment attachments, upload them, grade them and record results in a spreadsheet."

Developers and engineers will find coding improvements. OpenAI said GPT‑5.4 "incorporates the industry‑leading coding capabilities of GPT‑5.3‑Codex." Tomsguide reported that a Codex "fast mode" can deliver up to 1.5x speed gains and that an experimental Playwright (Interactive) feature lets Codex visually debug web and Electron apps in real time. The New Stack and other outlets said GPT‑5.4 matches or outperforms recent models on established coding benchmarks such as SWE‑Bench Pro.

Benchmarks cited in coverage painted a strong performance picture, though sources vary. Cryptobriefing reported that on the GDPval benchmark for professional knowledge work GPT‑5.4 matches or exceeds industry professionals in 83 percent of comparisons, versus roughly 71 percent for GPT‑5.2. The New Stack and Tomsguide reported wins over previous OpenAI Codex releases and competitor models on some coding and agentic tests.

OpenAI acknowledged a premium for the model in API pricing, saying the API price per token is higher than for GPT‑5.2 but that greater token efficiency reduces overall token use. The New Stack quoted OpenAI saying Batch and Flex pricing are available at half the standard API rate and that Priority processing is available at twice the standard API rate.

The release will accelerate automation of multi‑step professional workflows, from document and spreadsheet work to code testing and debugging. Observers noted commercial and reputational stakes. Gizmodo framed the launch as a bid by OpenAI to regain public favor amid recent controversies, an editorial assessment attributed to that outlet. OpenAI presented GPT‑5.4 as a step toward more capable, efficient tooling for professional users while marketplaces and regulators watch how broadly agentic automation changes workplace practices and security exposure.