OpenAI releases GPT-5.5, aiming to automate complex computer work

OpenAI is pushing GPT-5.5 as a model that can stay on a task longer and with less supervision, whether that means debugging code, refactoring a project, testing a fix, validating results or moving through browser-based work on local development servers and file-backed pages. The company said the system is designed to understand intent faster, plan across multiple steps, use tools, check its own work and keep going through ambiguity until the job is finished.

The immediate rollout to Plus, Pro, Business and Enterprise users in ChatGPT and Codex shows where OpenAI thinks the model belongs: inside daily workplace software, not just in chat windows. GPT-5.5 Pro is also rolling out to Pro, Business and Enterprise users in ChatGPT, while the API release will follow later after additional safeguards are put in place. OpenAI said the model is especially strong in agentic coding, computer use, knowledge work and early scientific research, the kinds of jobs that depend on sustained reasoning over time. Greg Brockman, OpenAI’s president, said the model can do more with less guidance.

Microsoft moved quickly as well. The company said GPT-5.5 will be generally available in Microsoft Foundry on April 24, a sign that enterprise distribution is becoming part of the product race itself. OpenAI said the model matches GPT-5.4’s per-token latency in real-world serving while operating at a much higher intelligence level, and that it uses fewer tokens to complete the same Codex tasks, a combination that matters for firms trying to automate software workflows at scale without driving up cost.

OpenAI is also trying to draw a line between capability and control. The company said GPT-5.5 went through its full safety and preparedness framework, with internal and external red-team testing and extra evaluation for advanced cybersecurity and biology capabilities. Nearly 200 early-access partners fed back on real use cases before launch. OpenAI said the system meets its High cybersecurity risk classification but not the more severe Critical threshold, a reminder that more autonomy does not erase the need for guardrails.

The release came less than two months after GPT-5.4, which OpenAI introduced on March 5 with a 1 million-token context window in Codex and the API, native computer-use capabilities and tool search. That earlier model was positioned as a major step toward state-of-the-art computer use, and GPT-5.5 extends the same direction with deeper long-context reasoning, more reliable agentic execution and greater token efficiency. The pace of releases, and the enterprise paths OpenAI and Microsoft are building around them, suggests the contest is shifting from flashy chat output to whether AI can become dependable workplace infrastructure across offices, codebases and regulated industries nationwide.