OpenAI Makes GPT-5.5 Instant the New Default ChatGPT Model, Cutting Hallucinations in Law and Medicine

OpenAI on May 5 swapped out the default brain behind ChatGPT, replacing GPT-5.3 Instant with a new model called GPT-5.5 Instant. The upgrade keeps the snappy response times users expect from the company's quick-reply tier while delivering meaningful jumps on math, multimodal reasoning, and—critically—reducing the kind of confident-but-wrong answers that have plagued AI assistants in regulated domains like law, medicine, and finance.

The benchmark gains are substantial. GPT-5.5 Instant scored 81.2 on AIME 2025, the high-school-olympiad math test that has become a standard reasoning yardstick, up from 65.4 for its predecessor. On the multimodal MMMU-Pro benchmark it climbed to 76 from 69.2. OpenAI says the model was specifically tuned to lower hallucination rates on "sensitive areas such as law, medicine, and finance," suggesting the company is positioning the default tier for higher-stakes professional use rather than just casual chat.

Beyond raw capability, the rollout brings a personalization layer. GPT-5.5 Instant can now reach into past conversations, uploaded files, and—if connected—Gmail to ground answers in a user's own context. Alongside the model, OpenAI is exposing memory sources across all models, letting users see which prior chats or documents shaped a response and delete or correct stale entries. Shared chats deliberately hide those memory sources so personal context doesn't leak when conversations are passed around.

Plus and Pro subscribers get the new model first on web, with mobile next, and OpenAI says Free, Go Business, and enterprise tiers will follow "in the coming weeks." Developers can hit it through the API as chat-latest; the outgoing GPT-5.3 Instant remains available to paid API users for three months. The launch lands alongside other ChatGPT updates this week, including a hands-free CarPlay integration and a new Interactive Learning module covering 70-plus math and science topics like the Pythagorean theorem and the ideal gas law.

Coming roughly two weeks after OpenAI's headline GPT-5.5 release pushed agentic coding capabilities forward, the Instant variant signals the company's strategy of pairing a high-end frontier tier with a fast, cheaper default model that quietly carries the bulk of consumer traffic. With rivals like Anthropic and Google racing to ship their own low-latency models, replacing the default is one of the highest-leverage moves OpenAI can make—every free-tier user gets the upgrade without changing a setting.