OpenAI has rolled out GPT-5.5 Instant as the new default model for ChatGPT. It replaces GPT-5.3 Instant and brings some notable improvements.
The numbers
The most important improvement is reliability: GPT-5.5 Instant produces 52.5 percent fewer hallucinated claims than its predecessor, according to OpenAI — tested on high-stakes prompts in medicine, law, and finance. On top of that, inaccurate claims dropped by 37.3 percent in particularly challenging conversations.
The model is also more concise: 30 percent fewer words and 29 percent fewer lines. Less verbosity is rarely bad news when it comes to language models.
New memory system
GPT-5.5 Instant can now use a search tool to reference past conversations, saved files, and connected Gmail to deliver more personalized responses. Plus and Pro users on the web got the feature first, with mobile coming next.
ChatGPT now also shows memory sources across all models — revealing which stored information influenced a response. You can delete outdated sources or correct them.
For developers
GPT-5.5 is available via the API as ‘chat-latest’. GPT-5.3 remains an option for paid users for three more months before being retired.
My take
The hallucination reduction sounds impressive, but benchmark numbers like these always deserve some caution — what matters is how it feels in daily use. I find the trend toward brevity more interesting. Language models that talk less and say more is a welcome shift across the entire industry.
Sources: OpenAI Blog · TechCrunch · Axios