OpenAI is giving ChatGPT a bigger job description. With GPT-5.5, the company says its newest model can take on more complicated work with less hand-holding from users.
OpenAI introduced GPT-5.5 as a “new class of intelligence for real work,” saying it’s capable of handling “messy, multi-part” tasks across coding, research, data analysis, and computer-based workflows.
More work, fewer nudges
OpenAI is not presenting GPT-5.5 as just another chatbot with better replies. The company described it as a model that can stay with a task longer, decide what needs to happen next, and carry more of the work without users breaking every step into tiny instructions.
According to the AI company, GPT-5.5 can “plan, use tools, check its work, navigate through ambiguity, and keep going.”
A user could bring it a rough problem instead of a polished prompt: a bug with an unclear cause, a pile of business inputs that needs structure, a research question with too many moving parts, or a workflow that crosses several tools. This means fewer nudges for users. GPT-5.5 can keep working once it understands the assignment, not simply waiting for the next command.
From bug hunt to code rewrite
Coding is where OpenAI makes its strongest case for GPT-5.5. The AI model is built for engineering work that does not end with one neat answer, from tracing a bug and understanding the system around it to making the fix and testing whether it holds.
The gains showed up in coding tests, including an 82.7% score on Terminal-Bench 2.0, which measures complex command-line work. More importantly for developers, GPT-5.5 is better at implementation, refactors, debugging, testing, and validation across a larger codebase.
Dan Shipper, founder and CEO of Every, called GPT-5.5 “the first coding model I’ve used that has serious conceptual clarity.” In one test, OpenAI said the model was given a broken post-launch app state and asked to produce the kind of rewrite an engineer had eventually chosen after days of debugging. GPT-5.4 couldn’t. GPT-5.5 could.
ChatGPT takes on the desk job
GPT-5.5 also reaches into the workday outside the code editor, including research, data analysis, business planning, and software-based tasks.
The model can take scattered inputs and turn them into usable output, whether that means sorting information, drafting a report, building a spreadsheet model, or moving through software to complete a task. In Codex, GPT-5.5 also outperforms GPT-5.4 at generating documents, spreadsheets, and slide presentations.
Inside OpenAI, the finance team used Codex to help review 24,771 K-1 tax forms across 71,637 pages, while another employee automated weekly business reports and saved five to 10 hours a week.
GPT-5.5 gets a guarded rollout
A model that can do more also creates a tougher access problem. OpenAI is adding stricter cyber controls around requests that could help users find, exploit, or misuse security weaknesses.
Some users may see more refusals at first as those systems are tuned. Verified cyber defenders, however, can apply for trusted access with fewer restrictions for legitimate security work.
GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. GPT-5.5 Pro is coming to Pro, Business, and Enterprise users, while API access is expected “very soon.”
Google is expanding its enterprise AI agent lineup as competition with OpenAI and Anthropic intensifies.