GPT-5.5 Instant Becomes ChatGPT's Default Model

OpenAI quietly swapped its default ChatGPT engine on May 5, replacing GPT-5.3 Instant with the newly released GPT-5.5 Instant. The change went out to all consumer users without much fanfare, though API customers got a more substantial upgrade: a one-million-token context window at $5 per million input tokens and $30 per million output tokens.

The headline claim is about hallucinations. In internal evaluations, OpenAI reports that GPT-5.5 Instant produced 52.5 percent fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in medicine, law and finance. On especially difficult conversations users had previously flagged for factual errors, inaccurate claims dropped by 37.3 percent.

If those numbers hold up outside the lab, that is a meaningful jump. Hallucination remains the persistent flaw of large language models, and we wrote recently about whether the problem is intrinsic to the architecture or simply a tuning issue. OpenAI is implicitly arguing the latter.

What changed

Beyond hallucination reduction, the company says GPT-5.5 Instant gives tighter answers across most subjects, with a more conversational tone and better use of personalisation context. TechCrunch reports the model is now ChatGPT's default for free, Plus and Pro tiers, replacing 5.3 Instant entirely.

For developers, the API listing is where the more interesting numbers live. The one-million-token context window puts GPT-5.5 in the same ballpark as Anthropic's Claude family and Google's Gemini Pro. Output is capped at the standard limits, but a million tokens of input means whole codebases, full court filings or several research papers can sit inside a single prompt.

Why it matters

The headline upgrade is the default change. Hundreds of millions of ChatGPT users will now be talking to GPT-5.5 without ever opting in or noticing the swap. If hallucination rates do drop by half on the kinds of questions people actually ask, the reputation problem that has dogged the chatbot since launch may finally start easing.

The pricing also matters. At $5 input and $30 output per million tokens, GPT-5.5 sits in a strange middle ground. It is cheaper than Claude Opus on input but pricier on output, and it is several times more expensive than the lighter Haiku and Flash tiers from rivals. OpenAI is not chasing the bottom of the market here. It is chasing trust.