BreakthroughsWednesday, May 6, 2026· 2 min read

ChatGPT’s new GPT-5.5 Instant dramatically cuts hallucinations

Source: The Verge AI

TL;DR

OpenAI says its new default ChatGPT model, GPT-5.5 Instant, produces far fewer hallucinated claims — especially on high-stakes topics like medicine, law, and finance. Internal tests report a 52.5% drop in hallucinations and a 37.3% reduction in inaccurate claims on user-flagged conversations, a meaningful step toward more reliable AI interactions for millions of users.

Key Takeaways

  • 1GPT-5.5 Instant is now the default model for ChatGPT and focuses on improved factuality.
  • 2OpenAI reports a 52.5% reduction in hallucinated claims on high-stakes prompts versus GPT-5.3 Instant.
  • 3The model also cut inaccurate claims by 37.3% in particularly challenging, user-flagged conversations.
  • 4Improvements come from internal evaluations; external testing and real-world use will further validate gains.
  • 5Fewer hallucinations mean safer, more trustworthy AI assistance across areas like healthcare, legal, and finance.

OpenAI’s GPT-5.5 Instant brings big gains in factual accuracy

OpenAI has rolled out GPT-5.5 Instant as the new default for ChatGPT, touting substantial reductions in hallucinations across a range of high-stakes topics. According to OpenAI’s internal evaluations, the model produced 52.5% fewer hallucinated claims on prompts covering areas such as medicine, law, and finance compared with its GPT-5.3 Instant predecessor.

The company also reports a 37.3% decline in inaccurate claims on especially challenging conversations that users had previously flagged for factual errors. Those improvements suggest the model is better at grounding responses and avoiding confidently stated mistakes — a key milestone for everyday reliability and user trust.

Why this matters: fewer hallucinations mean safer guidance and more dependable assistance for millions who rely on ChatGPT for research, professional queries, and decision support. Reductions on high-stakes topics are particularly encouraging because incorrect claims in those domains can have outsized consequences.

OpenAI’s figures come from internal testing, so independent assessments and real-world usage will help confirm the full scope of improvements. Still, making the model with stronger factuality the default for ChatGPT signals tangible progress toward AI that better supports users when accuracy matters most.

  • Default upgrade: GPT-5.5 Instant replaces the prior default in ChatGPT.
  • High-stakes win: 52.5% fewer hallucinations on critical prompts.
  • Flagged conversations: 37.3% fewer inaccurate claims on user-flagged exchanges.

Get AI Wins in Your Inbox

The best positive AI stories delivered to your inbox. No spam, unsubscribe anytime.