OpenAI releases GPT-5.5, built for coding and research work
OpenAI released GPT-5.5 to paying ChatGPT users and API developers seven weeks after launching GPT-5.4. The new model family prioritizes work tasks: coding, computer automation, and research.
The model scores higher than GPT-5.4 on benchmarks measuring ability to use applications across a computer and solve mathematical problems. It can work more independently than previous versions, completing tasks with less human direction.
What researchers need to know
GPT-5.5 is built as a general model, available to anyone. But researchers and people doing intensive computational work will likely find it most useful.
OpenAI President Greg Brockman said the model "can look at an unclear problem and figure out just what needs to happen next." He framed it as foundational to how AI agents will handle computer work at scale.
The company used GPT-5.5 and its Codex tool to help build the model during development, showing how AI now assists in its own creation. This reflects a broader shift: research itself increasingly relies on AI tools.
The security tradeoff
OpenAI implemented stronger cybersecurity safeguards in GPT-5.5 than previous versions. At launch, the model will be more cautious about requests involving cybersecurity work.
The company tested a specialized version, GPT-5.4-Cyber, with developers before any public release. This approach mirrors Anthropic's strategy with Claude Mythos Preview-letting experts find problems before broader access.
How researchers see their role changing
OpenAI's research leaders say capable AI models won't replace researchers. Instead, they expect humans to become orchestrators of AI-assisted research.
Jakub Pachocki, OpenAI's chief scientist, said the shift moves focus away from implementation details toward higher-level questions. "It allows you to make progress much more quickly and spend your focus, your energy on figuring out, 'What are the important things?'" he said.
Mia Glaese, vice president of research, said more capable models should raise "the threshold of what's worth building"-enabling research that wouldn't have been feasible before.
For researchers looking to integrate these tools into their work, AI Research Courses and AI Coding Courses cover practical applications in both domains.
Your membership also unlocks:
