OpenAI has launched GPT 5.4, positioning it as a faster, more capable AI model built for professional work across ChatGPT, the API, and Codex.
Quick Summary – TLDR:
- OpenAI released GPT 5.4 plus GPT 5.4 Thinking and GPT 5.4 Pro, aiming at professional tasks like coding, documents, and tool based workflows.
- The API version supports up to 1 million tokens of context, and OpenAI says the model is more token efficient than GPT 5.2.
- OpenAI reports stronger results across benchmarks for computer use, web navigation, and knowledge work, plus fewer factual errors.
- The launch also introduces Tool Search for cheaper tool calling, and new safety evaluation work around chain of thought monitoring.
What Happened?
OpenAI rolled out GPT 5.4 across ChatGPT, the API, and Codex, with two additional variants for different needs. The company says the model is its most capable and efficient option yet for professional work, with better coding, tool use, and reliability.
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. pic.twitter.com/1hy6xXLAmJ
GPT 5.4, Thinking, and Pro: What Each Version Is For
OpenAI is offering GPT 5.4 in three flavors. The standard GPT 5.4 is the main general model, while GPT 5.4 Thinking is positioned as the reasoning focused option inside ChatGPT. For users who want maximum performance, GPT 5.4 Pro is meant to push harder on complex tasks.
OpenAI also says GPT 5.4 pulls together advances in reasoning, coding, and agent workflows into one model, including coding strengths from its Codex line.
The Big Technical Upgrade: A 1 Million Token Context Window
For developers, one of the loudest announcements is the API context window that can reach 1 million tokens. That is aimed at long horizon work where an agent needs to read lots of material, plan steps, then execute and verify without losing track of earlier details.
OpenAI also highlighted improved token efficiency, claiming GPT 5.4 can solve similar problems using fewer tokens than GPT 5.2. For teams paying by usage, that can translate into lower total cost and faster completion.
Stronger Tool Use With Tool Search
OpenAI also reworked how GPT 5.4 handles tool calling in the API. Instead of stuffing every tool definition into the prompt up front, the company introduced Tool Search, which lets the model look up definitions only when needed.
That matters for agent systems that connect to lots of services. OpenAI says it can reduce token overhead, speed up requests, and make large tool ecosystems easier to run without losing intelligence.
Benchmarks and Real Work Claims
OpenAI is making a heavy benchmark case for GPT 5.4. It says the model set record results on OSWorld Verified and WebArena Verified, both focused on computer and browser use. On its own GDPval test for knowledge work tasks, OpenAI reports a record 83 percent score.
The company also pointed to results from Mercor’s APEX Agents benchmark for professional skills in law and finance. Mercor CEO Brendan Foody said:
Fewer Errors and New Chain of Thought Safety Checks
OpenAI says GPT 5.4 is 33-percent less likely to make errors in individual claims compared to GPT 5.2, and 18-percent less likely to include errors across full responses. The company also added a new evaluation focused on whether models can hide or misrepresent their chain of thought. OpenAI says deception is less likely in GPT 5.4 Thinking, suggesting chain of thought monitoring remains useful.
Availability Notes and the Bigger Context
GPT 5.4 is rolling out across ChatGPT and Codex, while the API offers GPT 5.4 and GPT 5.4 Pro. OpenAI also says GPT 5.4 Thinking replaces GPT 5.2 Thinking for paid users, with GPT 5.2 Thinking staying available for a limited time in a legacy section.
Some coverage around the launch also frames it as OpenAI trying to regain momentum after criticism tied to its Department of Defense work and reported user churn. OpenAI, however, is clearly betting that better accuracy, stronger coding, and more capable agents will pull attention back to the product.
SQ Magazine’s Takeaway
I see GPT 5.4 as OpenAI saying, we are done with small upgrades, here is the workhorse model for people who actually need results. The mix of native computer use, Tool Search, and a massive context window is not just nice to have, it is the kind of shift that makes agents feel practical instead of fragile. If the reliability gains are real, this is the kind of release that can make ChatGPT feel trustworthy again for serious work, not just quick answers.
