GPT-5.4 arrives with Thinking, Pro

OpenAI is turning its flagship model line into a clearer ladder of speed, cost, and reasoning depth rather than betting on one default model for everyone.

• –Base GPT-5.4 is positioned as the general professional-work model, with state-of-the-art coding and reasoning claims plus far cheaper pricing than Pro
• –GPT-5.4 Pro is clearly aimed at agentic and high-stakes workflows: it is slower, can take minutes on hard tasks, and is exposed in the Responses API for longer-running interactions
• –The 1.05M-token context window matters more than the branding bump because it makes large-repo analysis, long documents, and multi-step tool use much more practical
• –Splitting Thinking and Pro tiers gives developers a more explicit latency-versus-quality choice, but it also makes OpenAI’s pricing ladder much steeper for teams that want the best answers consistently