OpenAI · 2026-05-05 · major
GPT-5.5 Instant — New ChatGPT Default Cuts Hallucinations 52.5% on Medicine, Law, Finance Prompts
OpenAI rolled out GPT-5.5 Instant as the new ChatGPT default, replacing GPT-5.3 Instant. Internal evals show 52.5% fewer hallucinated claims on high-stakes medical, legal, and financial prompts and 37.3% fewer inaccuracies on user-flagged conversations.
OpenAI's new default ChatGPT model: more factual, more concise, less emoji noise.
Key specs
| Hallucination drop high stakes | -52.5% |
|---|---|
| Inaccuracy drop flagged convos | -37.3% |
| Aime2025 | 81.2 vs 65.4 |
| Mmmu pro | 76 vs 69.2 |
| Response length cut | -30.2% words |
What is it?
GPT-5.5 Instant is OpenAI's new default ChatGPT model and API entry, replacing GPT-5.3 Instant. It is the everyday low-latency tier — a separate product from the GPT-5.5 flagship reasoning model that shipped on April 23, 2026. Plus and Pro users get access on web first, with Free, Go, Business, and enterprise rollout in the following weeks.
How does it work?
OpenAI says the model was trained with sharper factuality objectives on medicine, law, and finance prompts where the prior default tripped up. It also picked up the GPT-5.5 family's tool-use behavior: it can call its search tool against your past conversations, files, and Gmail to ground replies, and a new memory sources panel shows which personal context shaped each answer. Stylistically the model is tuned to write 30.2% fewer words, fewer lines, and to drop gratuitous emojis.
Why does it matter?
ChatGPT's default model is what hundreds of millions of users actually hit every day, so a halving of hallucinations on high-stakes prompts is a much larger real-world delta than a flagship release that only paying power users see. The shift to grounded memory plus shorter answers is also a direct response to months of user complaints about verbosity and made-up citations.
Who is it for?
ChatGPT users on every plan; API developers using `chat-latest`.
Try it
API model id: chat-latest