OpenAI is rolling out GPT-5.4 Thinking and GPT-5.4 Pro in ChatGPT, while also making GPT-5.4 available in the API and Codex, according to posts from the company on X (OpenAI). The update frames GPT-5.4 as a single “frontier” model that combines improvements in reasoning, coding, and agentic workflows.
Alongside the ChatGPT rollout, OpenAI says the same model family is landing “gradually starting today” across ChatGPT, the API, and Codex, suggesting availability may vary by account or surface as the deployment progresses.
What OpenAI is emphasizing in GPT-5.4
OpenAI’s description focuses on a few developer-relevant themes rather than a long list of new features:
- Factuality and efficiency: GPT-5.4 is described as OpenAI’s “most factual and efficient” model, with fewer tokens and faster speed.
- One model across use cases: the company positions GPT-5.4 as bringing reasoning, coding, and agentic workflows together into a single model line, rather than splitting those capabilities across separate, specialized offerings.
That “fewer tokens” note is particularly notable in an API context, where output length can translate directly into latency and cost. OpenAI doesn’t provide benchmarks or pricing details in these posts, but it’s clearly pitching efficiency as a first-class improvement.
ChatGPT’s new “steering” behavior: interrupt and redirect mid-response
The most concrete user-facing change called out for ChatGPT is a new interaction pattern for GPT-5.4 Thinking: OpenAI says it’s now possible to interrupt the model and add instructions or adjust its direction mid-response.
OpenAI also ties Thinking mode to quality improvements when it spends more time: better context retention when it thinks for longer, plus improved deep web research.
As for availability, OpenAI says this “steering” capability is rolling out this week on Android and web, with iOS “coming soon.”
Rollout context: fast iteration, mixed reactions
The replies under OpenAI’s posts show a familiar split: excitement about coding gains and rollout access, alongside skepticism about reliability and broader product direction. A few comments also highlight how quickly model versions seem to be turning over—suggesting that, at least publicly, the cadence itself is becoming part of the story.
Still, the headline is straightforward: GPT-5.4 is now the active rollout across ChatGPT, the API, and Codex, with Thinking and Pro variants arriving gradually starting today, and a new mid-response steering interaction landing first on web and Android.