GPT-5.4

Available · Language, Multimodal, Reasoning

GPT-5.4 is an OpenAI reasoning model for text and image input, released on 5 March 2026 as GPT-5.4 Thinking in ChatGPT, as gpt-5.4 in the API and in Codex, alongside GPT-5.4 Pro. OpenAI calls it its first general-purpose model with native computer use and says it incorporates the coding capabilities of GPT-5.3-Codex. [1] [2]Model details Primary source

Timeline of GPT-5.4 →

Claims and evidence

  • Released Primary source[1]opening and Availability and pricing [3]Mar 5 (2026) [4]March 5, 2026: GPT-5.4 Thinking in ChatGPT
  • Status AvailableListed in the API model catalog and not on the deprecations page. Primary source[2] [5]
  • Successor of GPT-5.2GPT-5.4 Thinking replaced GPT-5.2 Thinking in ChatGPT; the system card notes there is no GPT-5.3 Thinking, so GPT-5.2 Thinking is the baseline. Primary source[1]Knowledge work; Availability and pricing [6]Introduction
  • Change · Tool use Adds native computer use: through the API's built-in computer tool it can issue mouse and keyboard commands in response to screenshots.Compared with GPT-5.2 Primary source[1]Computer use and vision [3]Mar 5 (2026)
  • Change · Context length Context window grows to 1,050,000 tokens in the API, from 400,000 for GPT-5.2.Compared with GPT-5.2 Primary source[2]Model details [7]Model details
  • Change · Tool use Adds tool search in the API, which gives the model a short tool list and loads a tool's full definition only when it is needed.Compared with GPT-5.2 Primary source[1]
  • Change · Safety OpenAI treats it as High cyber capability and deployed it with matching safeguards; the system card calls this a first for a general-purpose model.Compared with GPT-5.2 Primary source[6]
  • Input text, image Primary source[2]Model details
  • Output text Primary source[2]Model details
  • Feature Reasoning modeReasoning effort none (default), low, medium, high and xhigh. Primary source[2] [6]Introduction
  • Feature Computer useBuilt-in computer tool in the Responses API. Primary source[1]Computer use and vision [3]Mar 5 (2026)
  • Feature Function calling Primary source[2]Supported features
  • Feature Long context Primary source[3]Mar 5 (2026) [2]Model details
  • Context window 1,050,000 tokensThe announcement and changelog round this to 1M tokens; in Codex the standard window is 272K with experimental 1M support. Primary source[2]Model details
  • Parameters Not disclosedNot disclosed by OpenAI. No claim madeNo source
  • Access consumer app, API Primary source[1]Availability and pricing

Lineage

Predecessors

Successors

Based on

Not derived from another model.

Variants and derived

None recorded.

Siblings

None recorded.

All ancestors

All descendants

Variants

GPT-5.4 Pro

  • Released Primary source[1]opening and Availability and pricing [3]Mar 5 (2026)
  • Variant Inline variant in this record.Released the same day in ChatGPT (Pro and Enterprise plans) and in the Responses API as gpt-5.4-pro; OpenAI says it uses more compute to think harder. Primary source[1] [8]

Also known as: gpt-5.4-pro, gpt-5.4-pro-2026-03-05

Differs in:

  • Context window: 1,050,000 tokens Evidence not assessed [8]Model details

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from OpenAI on AI Radar →