OpenAI on May 5 released GPT-5.5 Instant as the new default model in ChatGPT, replacing GPT-5.3 Instant for all users. The headline claim, from OpenAI’s internal evaluations, is that the new model produces 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts spanning medicine, law, and finance.
The release is the latest in OpenAI’s “Instant” family, the lighter-weight, low-latency tier that powers most ChatGPT interactions. The lab framed the launch as a quality upgrade rather than a capability frontier extension — the focus is on smarter, more accurate answers and improved personalization controls inside the default model surface rather than new abilities for the long tail of complex tasks.
The May 5 announcement bundled several adjacent product moves. OpenAI introduced new realtime voice intelligence models in its API that the company said can reason, translate, and transcribe speech inside a single inference path. It opened a preview of a personal-finance experience inside ChatGPT for U.S. Pro subscribers, letting users connect accounts and see a money dashboard alongside the chat. It added Codex to the ChatGPT mobile app in preview, with remote SSH, hooks, access tokens, and HIPAA support extended for enterprise teams.
OpenAI did not publish full third-party benchmark scores for GPT-5.5 Instant on launch day, leaning instead on the hallucination-reduction figure and a high-level capability summary. The lab said the model would continue to roll out broadly across the ChatGPT user base as the default.
For users, the practical change is that ChatGPT’s default answer-generation should be measurably more reliable on questions where being wrong has real consequences. For OpenAI, the launch frames the lab’s competitive pitch in the next two quarters around trustworthy default behavior rather than new capability ceilings.