OpenAI just rolled out GPT-5.5 Instant as ChatGPT's new default model, and the headline feature is a major reduction in hallucinations. According to the company's internal evaluations, the model produces 52.5% fewer hallucinated claims than the previous GPT-5.3 Instant when handling high-stakes prompts in areas like medicine, law, and finance.
The improvements extend beyond just specialized domains. OpenAI says GPT-5.5 Instant reduced inaccurate claims by 37.3% on particularly challenging conversations that users had previously flagged for factual errors.
This matters because hallucinations remain one of the biggest barriers to using AI in professional settings. When you're drafting client communications, researching regulations, or analyzing data, you need to trust the output. A model that makes things up half as often is still not perfect, but it's a meaningful step toward reliability.
The catch is that these numbers come from OpenAI's own internal testing. Independent benchmarks will tell us how much this translates to real-world use. Still, any progress on factuality is worth paying attention to, especially if you're using ChatGPT for work where accuracy actually matters.
If you're a ChatGPT user, you're already using this model. OpenAI made it the default, so you don't need to switch anything manually.
