OpenAI deploys GPT‑5.5 Instant as ChatGPT's default

OpenAI rolled out GPT‑5.5 Instant as the default model for ChatGPT on May 5, 2026, replacing GPT‑5.3 Instant and pushing the new model into both the ChatGPT product and the API.

OpenAI and outside press framed the change as a measurable step forward on accuracy and context handling. The company said GPT‑5.5 Instant delivers fewer hallucinations and clearer use of prior conversation turns, and independent outlets reported the rollout began on May 5.

OpenAI highlighted a roughly 52.5% reduction in hallucinated claims on high‑stakes prompts covering medicine, law and finance — a figure cited by coverage summarizing OpenAI’s release notes and internal evaluations. Those reductions are presented as central to making the model safer for professional workflows.

The company also said GPT‑5.5 Instant is available in the API as the chat‑latest endpoint, which lets developers route new and existing ChatGPT traffic to the updated model without changing client code. OpenAI’s developer documentation and release notes list the model family and the relevant API names.

GPT‑5.5 sits alongside other variants in the release cadence: Instant (fast, concise), Thinking (deeper reasoning for paid tiers), and Pro for higher‑capacity tasks. OpenAI positioned the family as a coordinated set of capabilities for consumers, professionals, and platform partners.

OpenAI described extensive safety and governance testing for the GPT‑5.5 family, including domain‑specific evaluations and new targeted checks for biology and cybersecurity risks. The company said it deployed stricter classifiers for potential cyber misuse as part of the rollout.

The release is explicitly pitched at enterprise customers who use ChatGPT for tasks that depend on factual accuracy — compliance checks, customer support, and research synthesis among them. OpenAI and outside reports argue the accuracy gains make the service more reliable for these workflows, though businesses will still need their own validation processes.

OpenAI also announced new model‑adjacent tools for scientific and life‑sciences workflows, such as GPT‑Rosalind, intended to speed genomics and drug‑discovery reasoning. The company said these are part of a broader effort to specialize models for high‑stakes professional domains.

Early reviews and hands‑on articles from outlets testing ChatGPT reported shorter, more focused answers and less overexplaining in everyday prompts after the switch to GPT‑5.5 Instant, echoing OpenAI’s aim for a conciser conversational tone. Reporters noted the shift was noticeable in both free and paid tiers.

For developers, the immediate practical effect is that default ChatGPT behavior and the chat‑latest API responses will trend toward GPT‑5.5 Instant’s tradeoffs — faster replies and tighter context use. OpenAI warns that some enterprises may need to retune prompt designs and testing suites to match the new model’s behavior.

What to watch next: adoption in regulated fields, the real‑world size of accuracy improvements outside benchmark tests, and how rapidly OpenAI updates safety controls as specialized use cases emerge. The company’s data and third‑party audits will be central to verifying the claimed reductions in factual errors.