GPT-5.4: OpenAI's Frontier Model for Professional Work with Native Computer Use, Tool Search and 1M Context

What GPT-5.4 is

GPT-5.4 is a language model from OpenAI, released on 5 March 2026 in ChatGPT as GPT-5.4 Thinking, in the API as gpt-5.4 and in Codex. OpenAI called it its most capable and efficient frontier model for professional work. It brings together OpenAI's recent advances in reasoning, coding and agentic workflows in a single model, incorporating the coding capabilities of GPT-5.3-Codex while improving work across tools, software environments, spreadsheets, presentations and documents. OpenAI released GPT-5.4 Pro alongside it for maximum performance on complex tasks. On LMArena's text leaderboard GPT-5.4 has 1465 Elo (rank 57).

Specifications

In the API GPT-5.4 has a 1.05M-token context window, up to 128,000 output tokens and a knowledge cutoff of 31 August 2025. It takes text and image input and returns text; audio and video are not supported. Reasoning effort can be set to none, which is the default, low, medium, high or xhigh. OpenAI lists support for computer use, MCP, tool search and skills. Pricing is $2.50 per million input tokens and $15 per million output tokens, with cached input at a tenth of the input rate. For GPT-5.4 and GPT-5.4 Pro, prompts above 272K tokens are billed at a higher rate, and regional data-residency endpoints carry a 10% uplift.

Computer use, tools and efficiency

OpenAI says GPT-5.4 is the first general-purpose model it has released with native computer-use capabilities, able to write code that operates computers through libraries such as Playwright and to issue mouse and keyboard commands from screenshots. Developers can steer its behaviour through developer messages, including custom confirmation policies for different levels of risk. Tool search helps agents find and use the right tools across large ecosystems of tools and connectors. OpenAI also calls GPT-5.4 its most token-efficient reasoning model at launch, using significantly fewer tokens than GPT-5.2 to solve problems. In ChatGPT, GPT-5.4 Thinking can show an upfront plan of its thinking that users can adjust mid-response.

Results reported by OpenAI

OpenAI reports 83.0% on GDPval, which tests well-specified knowledge work across 44 occupations, against 70.9% for GPT-5.2. It reports 57.7% on SWE-Bench Pro, 75.0% on OSWorld-Verified, above the 72.4% human performance OpenAI cites and GPT-5.2's 47.3%, 54.6% on Toolathlon and 82.7% on BrowseComp. On an internal set of spreadsheet modelling tasks of the kind a junior investment banking analyst might do, it scored 87.3% against 68.4% for GPT-5.2. OpenAI says GPT-5.4's individual claims were 33% less likely to be false than GPT-5.2's on prompts where users had flagged factual errors. These are OpenAI's own figures.

Availability and successors

GPT-5.4 Thinking replaced GPT-5.2 Thinking in ChatGPT for Plus, Team and Pro users, with GPT-5.2 Thinking kept under Legacy Models until its retirement on 5 June 2026, and Enterprise and Edu admins could enable early access. GPT-5.4 Pro went to Pro and Enterprise plans. OpenAI said the name reflected the jump of folding GPT-5.3-Codex's coding capabilities into a mainline reasoning model. GPT-5.4 mini and nano followed on 17 March 2026, and OpenAI has since released the GPT-5.5, GPT-5.6 and GPT-6 generations.

Official Resources