ChatGPT 5.5 Is All About Math, Science and AI Research

Exactly seven weeks after releasing ChatGPT 5.4, OpenAI is dropping a new family of models. ChatGPT 5.5 is now available for paying ChatGPT and Codex users, along with the thinking and pro versions, with plans to bring it to the API soon. These models are built for work — specifically, coding, computer use and research.

OpenAI President Greg Brockman said GPT-5.5 is more intuitive than previous models and can do more with less human guidance.

“It can look at an unclear problem and figure out just what needs to happen next,” Brockman told reporters in a briefing. “It really feels like it’s setting the foundation for how we’re going to do computer work going forward, or how agent computing at scale will work.”

This is a general model, so anyone can use it. But it’s likely going to be the most useful for people doing research or other intensive tasks, like coding. It has agentic capabilities, which means it can independently complete tasks. It scored higher than GPT-5.4 on benchmarks that measure a model’s ability to use apps across your computer and solve math problems. 

OpenAI and other tech companies are trying to build models that act like true digital assistants, managing notifications and overseeing projects across your entire computer, not just one program. GPT 5.5 is the company’s biggest swing at this yet.

The company says GPT-5.5 has its strongest cybersecurity safeguards yet, and at launch, it will be more conservative in the cybersecurity requests it allows. This is because of growing concerns that increasingly cyber-capable AI models can find weaknesses in our existing internet structure. That’s why both Anthropic’s Claude Mythos Preview and OpenAI’s 5.4-Cyber were released to developers to root out problems before a potential public release.

AI in research

For OpenAI, research is a two-way street. GPT 5.5 is part of the foundation of the super app that the company aims to build out of Codex. Research on AI is also increasingly done with the help of AI tools; Brockman said the company used 5.5 (and Codex) to help build itself during development.

AI has changed how research is done over the past few years, and the field isn’t immune to the concerns of other industries that AI may automate humans out of jobs in the future. While GPT-5.5 is OpenAI’s most capable research model yet, the company’s top scientists and researchers aren’t worried about being replaced by AI.

Mark Chen, OpenAI’s chief research officer, discussed pursuing a near-term goal of having humans serve as the “orchestrators” of research heavily assisted by AI. 

Mia Glaese, vice president of research, said she was hopeful that more capable AI models would enable more research and raise “the threshold of what’s worth building.”

“I feel more productive because the challenge transitions from figuring out the details of the implementation, the low-level abstractions, to the higher-level goals,” said Jakub Pachocki, OpenAI’s chief scientist. “It allows you to make progress much more quickly and spend your focus, your energy on figuring out, ‘What are the important things?’…There is no point in automating things for their own sake.”