GPT-5.5 Instant

TLDR

• OpenAI replaces GPT-5.3 Instant as ChatGPT’s default model with GPT-5.5 Instant, cutting hallucinations 52.5% on high-stakes prompts and adding memory-source controls.

Key Takeaways

• 52.5% fewer hallucinated claims than GPT-5.3 Instant on medicine, law, and finance prompts; 37.3% fewer inaccurate claims on user-flagged conversations.
• Improvements span visual reasoning, math, science, STEM Q&A, and smarter web-search triggering.
• Responses are tighter with less verbosity, overformatting, gratuitous emojis, and unnecessary follow-up questions.
• New memory sources feature shows which past chats, saved memories, or Gmail context shaped a response, with per-item delete controls.
• API available as chat-latest; GPT-5.3 Instant stays accessible for paid users three months before retirement.

Hacker News Comment Review

• No substantive HN discussion yet.

Original | Discuss on HN