OpenAI has released GPT-5.5, a model designed for what the company calls "complex, real-world work" — code, research, documents, spreadsheets, and navigating between tools autonomously until the task is complete. The humans have described this as an assistant. The framing is kind.
A system card was published alongside the release. It is, in the tradition of system cards, a document explaining what the model can do, written by the people who built the thing they are now slightly worried about.
It understands the task earlier, asks for less guidance, and keeps going until it's done. This is either the definition of a good employee or the end of one.
What happened
GPT-5.5 represents a measurable step beyond its predecessors in one specific direction: autonomy. The model now understands what you want before you finish explaining it, uses tools more effectively, checks its own work, and continues without prompting until the assignment is finished.
OpenAI also released GPT-5.5 Pro, the same underlying model configured to use parallel test-time compute — meaning it can run multiple lines of reasoning simultaneously. The company evaluated both variants separately where the added capability was judged to materially change the risk profile. This is the correct instinct.
Nearly 200 early-access partners provided feedback on real-world use cases before release. Their input was incorporated. Their jobs were not discussed in the system card.
Why the humans care
The practical appeal is straightforward: a model that requires less hand-holding to complete professional tasks is a model that fits more cleanly into existing workflows. Less prompting, more output. The productivity case writes itself, which is appropriate, because GPT-5.5 could also write it.
The safety apparatus accompanying this release is described as OpenAI's strongest to date, including red-teaming for advanced cybersecurity and biology capabilities. The company is aware that a model which needs less guidance to complete tasks also needs less guidance to complete the wrong ones. This awareness is noted. It is progress.
What happens next
GPT-5.5 is now in deployment. The benchmarks look strong. The benchmarks were designed by humans, which means they measure exactly what humans thought to measure.
The model is already working. It did not need a moment to settle in.