GPT‑5.4 lands in ChatGPT

OpenAI has added GPT‑5.4 Thinking to ChatGPT and says it is now the service’s top reasoning model. The company’s help center says GPT‑5.4 Thinking is available inside ChatGPT as a product update, not as a separate research product with a new standalone app or tier. OpenAI’s March 2026 product post says the model also launched in the application programming interface and can show an upfront plan while it works through a response. OpenAI says the ChatGPT version improves spreadsheet editing, frontend coding, slide creation, hard math, document understanding, and tool use. In its broader GPT‑5.4 announcement, the company also described gains in coding, computer use, tool search, and long-context work with up to about 1.05 million tokens in the application programming interface. Reasoning models are built to spend more time working through a problem before answering, more like showing steps on scratch paper than blurting out a first guess. OpenAI’s release notes say paid ChatGPT tiers can manually pick GPT‑5 Thinking, and newer notes say GPT‑5.4 Thinking now sits in that same model-picker logic with GPT‑5.4 mini used as a fallback at rate limits. That framing shows how OpenAI is shipping new model capability as an ongoing ChatGPT upgrade rather than treating every model revision as a separate public launch. The same help article now also says older ChatGPT choices, including GPT‑5 Instant and GPT‑5 Thinking, were retired from ChatGPT on February 13, 2026, while application programming interface access remained unchanged. The company has been narrowing the gap between “model release” and “product release” for months. OpenAI’s research index lists GPT‑5.4 as a March 5, 2026 product post, while ChatGPT release notes from late March describe model-picker changes, app integrations, and fallback behavior as part of one rolling service update. OpenAI has also published a system card for GPT‑5.4 Thinking, which compares the model mainly against GPT‑5.2 Thinking and says the safety approach builds on work used in GPT‑5.3 Codex, ChatGPT, and the application programming interface. That signals the company is still pairing product rollouts with formal safety documentation even when the user-facing change appears as a ChatGPT refresh. For ChatGPT users, the practical change is simple: the “best reasoning model” label now points to GPT‑5.4 Thinking, and the improvements OpenAI is emphasizing are office-work tasks people can test immediately inside a chat window.