OpenAI pushes GPT-5.5 toward workflows

OpenAI rolled out GPT-5.5 on April 23 and described it less as a chatbot than as software for getting work done on a computer. (openai.com) In its launch post, OpenAI said GPT-5.5 can take “messy, multi-part” requests, plan steps, use tools, check results and keep going across software until a task is finished. (openai.com) The company released GPT-5.5 to Plus, Pro, Business and Enterprise users in ChatGPT and Codex on April 23, then added GPT-5.5 and GPT-5.5 Pro to the application programming interface on April 24. (openai.com 1) (openai.com 2) OpenAI tied the launch to coding first. In the Codex changelog, it called GPT-5.5 the recommended model for implementation, refactors, debugging, testing, validation and “knowledge-work artifacts.” (developers.openai.com) The underlying idea is simple: instead of answering one prompt at a time, the model is supposed to carry a job across several steps, the way a worker moves from request to draft to verification. OpenAI said GPT-5.5 is built for coding, online research, data analysis, documents, spreadsheets and software operation. (openai.com) (help.openai.com) OpenAI’s benchmark table framed that shift as measurable progress in “agentic” work, or tasks that require planning and tool use over time. GPT-5.5 scored 82.7% on Terminal-Bench 2.0 versus 75.1% for GPT-5.4, 78.7% on OSWorld-Verified versus 75.0%, and 84.4% on BrowseComp versus 82.7%. (openai.com) The company also said GPT-5.5 matched GPT-5.4 per-token latency in real-world serving while using “significantly fewer tokens” to finish the same Codex tasks. That pairs a higher-capability pitch with a cost-and-speed pitch for developers and enterprise buyers. (openai.com) OpenAI’s release notes put the same point in product terms. They said GPT-5.5 performs better on complex terminal workflows, real-world GitHub issue resolution, long-horizon coding tasks, search, retrieval and document-grounded question answering across reports and policies. (help.openai.com) The company paired the rollout with a heavier safety message than in a typical model launch. OpenAI said it ran GPT-5.5 through its Preparedness Framework, did targeted red-teaming for cybersecurity and biology risks, and gathered feedback from nearly 200 early-access partners. (openai.com 1) (openai.com 2) Codex updates around the launch show where OpenAI thinks this goes next. The app now includes browser use and computer use features so Codex can click through rendered pages, verify fixes and operate macOS apps in some regions. (developers.openai.com) The release leaves OpenAI selling less conversation and more completion. GPT-5.5 is being positioned as the model that takes a rough instruction and pushes it through the workflow until there is something finished to review. (openai.com)