OpenAI rolls out GPT-5.5

OpenAI changed the default engine inside ChatGPT again — and this one matters more than a routine model swap. On May 5, the company began rolling out GPT-5.5 Instant as the standard model for logged-in ChatGPT users, while also pointing the API alias `chat-latest` at the same model. OpenAI says the pitch is simple: faster answers, fewer made-up claims, and better everyday performance. But the bigger story is that the “default model” is becoming a moving target. (openai.com) ### What actually shipped? GPT-5.5 Instant is not the full GPT-5.5 model that OpenAI introduced on April 23. It is the lighter, faster “Instant” version — the workhorse meant for normal ChatGPT use and for developers who want the latest default behavior through `chat-latest`. In ChatGPT, it replaces GPT-5.3 Instant as the default for logged-in users. In the API, OpenAI’(openai.com)del family developers should target. (openai.com) ### Why is “Instant” the point? Because most people do not want the smartest possible model at any cost. They want the one that feels quick, steady, and good enough across lots of tasks. OpenAI is positioning GPT-5.5 Instant as that middle lane — better at info-seeking questions, how-tos, technical writing, translation, image understanding, STEM questions, and decid(openai.com)t assistant model, not the prestige model. (help.openai.com) ### How much better is it supposed to be? The headline number is hallucinations. OpenAI says GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in medicine, law, and finance. It also says the model reduced inaccurate claims by 37.3% on especially hard conversations that users had flag(help.openai.com)spel — but they tell you what OpenAI thinks the upgrade is for. (openai.com) ### Can users stay on the old model? For a while, yes. Paid users can still access GPT-5.3 Instant through model settings for three months before OpenAI retires it. That is a real clue about how the company expects these transitions to go now. The default moves first. The fallback survives briefly. Then the old model disappears. (openai.com)e `chat-latest` now means GPT-5.5 Instant. If your app depends on the alias instead of pinning a specific model, your behavior can change without a code rewrite. Sometimes that is great — you get better quality automatically. But the catch is that prompts, tool use, latency, and edge-case behavior can all shift at once. OpenAI’s pro(openai.com) prompts often work better than older, process-heavy prompt stacks. (openai.com) ### Does this change the way teams build on APIs? Yes — a little, but in an important way. Vendor model releases used to feel like long-lived platforms. Now they look more like fast-moving software dependencies. If the default can change in weeks, teams need their own evals, regression tests, and rollback plans. You cannot just trust that “latest” will stay close enough(openai.com)across AI platforms. OpenAI is making it explicit. (openai.com) ### How does this fit the bigger GPT-5.5 rollout? It completes the stack. OpenAI launched full GPT-5.5 and GPT-5.5 Pro in late April for heavier-duty work like coding, research, data analysis, and agentic tasks. Now the company has pushed the lighter Instant version into the default consumer slot. So the message is not just “here is a new flagship.” It is “the whole product line is moving.” (openai.com) ### Bottom line? GPT-5.5 Instant is a product update, but it is also a policy signal. OpenAI wants the default assistant to improve continuously, even if that means the floor keeps moving under users and developers. For everyday ChatGPT users, that probably means better answers with less fuss. For builders, it means “latest” is convenient — but only if your own testing is strong enough to survive it. (openai.com)