OpenAI on May 5 began rolling out GPT-5.5 Instant to all ChatGPT users, replacing GPT-5.3 Instant as the default model in ChatGPT and as the chat-latest alias in the API. The model trims response length, cuts hallucinated claims, and surfaces memory sources users can inspect, edit, or delete.

Try it: ship a tighter ChatGPT brief in 30 minutes

Open ChatGPT, confirm the model picker now reads "GPT-5.5 Instant" by default, and re-run a brief you wrote against GPT-5.3 Instant a week ago. Compare word count and tone: OpenAI says GPT-5.5 Instant produces 30.2% fewer words and 29.2% fewer lines on average, which usually means tighter creative briefs, shorter outline drafts, and cleaner email replies without the boilerplate hedging the older model added. If you ship code or content through the API, swap your Instant deployments to the chat-latest alias, then re-test factual prompts (legal disclaimers, medical copy, financial figures) against your evals before pushing to production.

Why It Matters

The default model is the one most ChatGPT users never change, and it powers the bulk of free-tier and Plus prompts that creators run for ideation, scripting, and rewriting. OpenAI claims a 52.5% reduction in hallucinated claims on high-stakes topics like medicine, law, and finance versus GPT-5.3 Instant, plus a 37.3% drop in inaccurate claims on conversations users had flagged as factually wrong. TechCrunch notes the model also keeps the low latency of its predecessor, so the swap is invisible for casual prompts but meaningfully more reliable for client-facing copy.

Key Details

GPT-5.5 Instant lands as a sibling to the GPT-5.5 thinking model OpenAI launched in late April for harder reasoning workloads. The Instant variant is the latency-first cousin: shorter answers, faster turns, and now better grounding on photo and image uploads, STEM questions, and decisions about when to invoke web search. The system card documents the eval methodology behind the hallucination numbers.

Personalization is the second headline. Instant now pulls context from past chats, uploaded files, and Gmail (when connected) and shows a per-message memory source list users can review, correct, or delete. Shared chats hide the source trail, so handing a transcript to a collaborator no longer leaks personal context.

What to Do Next

If you run a content workflow on GPT-5.3 Instant via the API, treat May 5 as your soft migration day: pin to the explicit gpt-5.3-instant string for three months if you need stability, and stage chat-latest behind a feature flag while you re-baseline your evals on the new tone and length distribution. For studio teams, audit your shared-chat workflow once: confirm clients on shared links cannot see your private memory sources before the next brief goes out.

How to integrate this

For working creators using ChatGPT in production: GPT-5.5 Instant becomes the default automatically when this rolls out, but you can still pin GPT-5.3 Instant in the model selector if you have specific prompts tuned for the prior model. Most creator workflows benefit from the new default without changes; if your existing prompts reference model-specific quirks, test 3-5 representative prompts before committing.

Frequently asked questions

What changed in GPT-5.5 Instant compared to GPT-5.3?

GPT-5.5 Instant is OpenAI's new default ChatGPT model, replacing GPT-5.3 Instant. The change improves response speed and integrates better with ChatGPT's memory and source-citation features. Most creator workflows benefit without prompt changes.

Can I still use GPT-5.3 Instant?

Yes, GPT-5.3 Instant remains available via the model selector for users with specific prompts tuned to the prior model's behavior. The default just shifts to 5.5 unless you pin 5.3.

Does this affect ChatGPT API pricing?

Pricing for the chat API endpoints follows OpenAI's standard model-tier pricing. Check platform.openai.com/docs/pricing for current rates. The default-model swap does not change billing structure for existing API users.