GPT-5.5 arrived April 23, and OpenAI is billing it as a model that can take on ambiguous problems without being walked through them step by step. Less hand-holding, more initiative. President Greg Brockman put it plainly at a press briefing: the model looks at an unclear situation and works out what needs doing next. His read is that it points toward how people will interact with computers going forward.
The release comes roughly seven weeks after GPT-5.4 shipped, which says something about the rhythm OpenAI is keeping right now.
Where the model is meant to earn its place is in technical and analytical work: reading data, writing and fixing code, operating software directly, doing online research, producing documents and spreadsheets.
OpenAI also released a safety classification for it. It doesn’t reach the company’s “Critical” cybersecurity threshold, the one associated with entirely new pathways to serious harm, but it does meet the “High” bar, meaning it can amplify existing ones. Mia Glaese, VP of research, said the model went through third-party safeguard testing and red-team exercises covering cyber and biological risks, and that OpenAI has been refining those safeguards over months as its models have grown more capable in that domain.
The cybersecurity question has sharpened considerably since Anthropic revealed its Claude Mythos model earlier this month. That release came with restrictions on deployment, partly because of how well the model could identify software vulnerabilities. Wall Street has been paying close attention.
GPT-5.5 is available from April 23 to paying ChatGPT subscribers across Plus, Pro, Business, and Enterprise tiers, and through Codex, the company’s coding assistant. API access is coming, OpenAI says, though it will need its own set of safeguards before it goes live.