OpenAI unveils ChatGPT Work and GPT-5.6 AI models to automate complex enterprise tasks

OpenAI has launched ChatGPT Work, an AI agent capable of completing multi-step projects across apps and files, alongside the GPT-5.6 model family, introducing enhanced reasoning, coding, automation and productivity capabilities for businesses and developers.

OpenAI has introduced ChatGPT Work, a new AI agent within ChatGPT designed to execute complex, multi-step tasks across applications, files and workflows. Powered by the newly announced GPT-5.6 model family, the agent can independently gather context, work on projects for extended periods and generate completed outputs such as documents, spreadsheets, presentations and web applications.

The launch also expands the capabilities of ChatGPT across web, desktop and mobile platforms, positioning the platform as a unified productivity workspace for enterprises, professionals and developers.

Unified workspace integrates chat, work and codex

As part of the rollout, OpenAI has introduced a redesigned desktop application that combines Chat, Work and Codex into a single interface. While Chat continues to serve as an AI assistant for everyday tasks, Work is aimed at handling complex projects requiring multiple steps, and Codex remains focused on software development and technical workflows.

The desktop application also introduces a unified plugins directory that connects ChatGPT with business applications such as Slack, Gmail, Google Drive, calendars and customer relationship management (CRM) platforms. Additional capabilities include Scheduled Tasks for recurring workflows, an integrated browser and Computer Use functionality that allows deeper interaction with websites and locally installed applications.

GPT-5.6 family expands AI performance across workloads

OpenAI also announced the GPT-5.6 family of models comprising Sol, Terra and Luna, which will be available across ChatGPT, Codex and the company's API offerings.

According to the company, GPT-5.6 Sol is its flagship model, designed to improve intelligence and efficiency across coding, cybersecurity, scientific research and enterprise knowledge work while reducing token usage and operating costs. Terra is positioned as a balanced model for everyday business tasks, while Luna is intended as the company's most cost-efficient option.

The GPT-5.6 models also introduce improvements in multi-step reasoning, template-based document creation, presentation and spreadsheet generation, as well as enhanced computer-use capabilities. OpenAI has also launched Ultra Mode, a high-performance setting that enables multiple AI agents to coordinate in parallel for demanding workloads.

Global rollout begins across platforms

OpenAI said Chat, Work and Codex are rolling out globally on Mac and Windows desktop applications for all users, including those on the Free tier. ChatGPT Work will initially become available on web and mobile for Pro, Enterprise and Edu customers, with availability for Plus and Business subscribers following shortly.

GPT-5.6 Sol will be accessible through medium and higher reasoning settings for Plus, Pro, Business and Enterprise users, while Pro and Enterprise customers will also have access to GPT-5.6 Pro for more computationally intensive tasks.

OpenAI highlights benchmark gains

The company said GPT-5.6 delivers significant performance improvements across several AI benchmarks.

According to OpenAI, GPT-5.6 Sol achieved a score of 53.6 on Agents' Last Exam, outperforming competing models while requiring fewer tokens and lower computational cost. On the Artificial Analysis Coding Agent Index, Sol recorded 80.0, surpassing previous benchmark leaders with reduced latency and token usage.

OpenAI also reported improvements in cybersecurity evaluations, with GPT-5.6 Sol scoring 73.5% on ExploitBench compared with 47.9% for GPT-5.5. The company added that the new model family also demonstrated stronger performance in coding-agent tasks and long-running professional workflows while delivering lower estimated API costs than competing models.