← Back
OpenAI
OpenAI rolls out GPT-5.4 mini and updates reasoning models with mid-response course correction
OpenAI APIChatGPTOpenAI · releasefeaturemodelapi · help.openai.com ↗

GPT-5.4 Mini and Thinking Models Now Available

OpenAI is expanding its ChatGPT model lineup with GPT-5.4 mini, which is now rolling out to users. Free and Go users can access GPT-5.4 mini through the "Thinking" feature in the menu. For Plus, Pro, and other paid subscribers, GPT-5.4 mini serves as a rate-limit fallback when GPT-5.4 Thinking reaches capacity, ensuring continued access to reasoning capabilities during peak usage. Enterprise customers have the option to default to GPT-5.4 mini auto-routing if preferred.

GPT-5.4 Thinking capabilities: The flagship GPT-5.4 Thinking model brings together reasoning, coding, and agentic workflows. Key improvements include:

  • Mid-response adjustment: The model now provides an upfront plan of its thinking, allowing users to adjust course while it's working and refine outputs without additional turns
  • Enhanced web research: Improved deep research capabilities, particularly for highly specific queries
  • Better context management: Improved ability to maintain context for longer conversations with extended thinking windows
  • Faster delivery: Higher-quality answers arrive faster while staying relevant to the task

Model Updates and Deprecations

GPT-5.3 Instant refinements: OpenAI has rolled out updates to GPT-5.3 Instant focusing on conversational quality. The model now reduces teaser-style phrasing ("If you want…", "You'll never believe…") and improves follow-up tone. Earlier updates to GPT-5.3 Instant emphasized more measured, grounded responses and clearer answers for advice-seeking questions.

GPT-5.3-Codex launch: The new GPT-5.3-Codex model combines Codex and GPT-5 training stacks for agentic coding capabilities. It's approximately 25% faster than previous versions, sets new benchmark highs, and enables general-purpose coding agent workflows that users can actively steer during execution.

Model retirements: As of March 11, 2026, GPT-5.1 models (Instant, Thinking, and Pro) are no longer available in ChatGPT. Existing conversations automatically migrate to their current equivalents (GPT-5.3 Instant, GPT-5.4 Thinking, or GPT-5.4 Pro). Earlier in February, GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini were retired from ChatGPT, though API availability remains unchanged.

Developer and User Actions

GPT-5.4 mini will not appear as a selectable model in the model picker; it functions only as an automatic fallback. GPT-5 Thinking mini will be retired as a selectable option in 30 days. Users and developers should update model selection preferences where applicable and review the official blog post and help center documentation for migration guidance on deprecated models.