GPT-5.5 Instant: OpenAI's faster, more accurate default model

Key Takeaways

• New default for everyone. GPT-5.5 Instant is OpenAI's daily-driver model and now replaces GPT-5.3 Instant as the default for all ChatGPT users. Note: this is not the full GPT-5.5 flagship — it's the lightweight, high-volume default.
• Far fewer hallucinations. OpenAI reports 52.5% fewer hallucinated claims on high-stakes prompts (medicine, law, finance) and 37.3% fewer inaccurate claims on conversations users had flagged for factual errors.
• Tighter, cheaper answers. Responses use 30.2% fewer words on average for the same quality — directly reducing latency and token overhead when the model runs inside an agent loop.
• Smarter routing and context. Better image/STEM reasoning, smarter about when to pull from the web, and stronger personalization from past chats, uploaded files, and connected apps like Gmail. A new "memory sources" feature shows exactly which past chats or memories shaped a response, so you can delete stale ones.
• Use it as the speed layer. The honest framing from the video: route quick tasks, light sub-agent delegation, and fast output checks to GPT-5.5 Instant — but keep a heavier frontier model for complex multi-step builds.
• API access. On the OpenAI API it's available as chat-latest, the model string that always points to the current default. Reportedly a stopgap ahead of an unannounced GPT-5.6.

Commands & Code Mentioned

hermes profile create gpt-5.5-instant   # create a dedicated profile, then authenticate via OpenAI OAuth (no separate API key)

chat-latest                              # OpenAI API model string that always points to the current default model