Bottom line

OpenAI's verified profile lists a July 9, 2026 release date for ChatGPT Work, an agentic layer built into the ChatGPT web, mobile, and desktop applications. Powered by the GPT-5.6 model family and Codex, it is designed to turn high-level goals into finished artifacts by gathering context from explicitly connected apps and workflows over hours-long, multi-step projects.

Signal

ChatGPT Work represents a formal push into long-running agentic workflows. Instead of acting purely as a synchronous conversational assistant, Work breaks complex projects into smaller steps and can stay active on a task for hours. Primary materials confirm it generates finished work artifacts, such as documents, spreadsheets, presentations, reports, and web applications.

The agent draws its context by connecting to specific application and file ecosystems, such as Slack, email, drives, calendars, and CRMs. It is available across web, mobile, and desktop (Mac and Windows). Access rules and capabilities are strictly tied to subscription plans: Free and Go users access the GPT-5.6 Terra model, while Plus, Pro, Business, and Enterprise users can choose between GPT-5.6 Sol, Terra, and Luna, with the added ability to set a specific "effort level" for execution.

What is not settled

The exact implementation boundaries of ChatGPT Work remain unresolved. The verified profile frames it as an agent gathering context across explicitly connected apps, files, and workflows—not as a system with deep OS-level integration. Assumptions regarding API-level access to enterprise data silos, generalized screen-scraping capabilities, or unfettered access to local files are unsupported by current primary sources.

Additionally, the exact context window size for ChatGPT Work is not settled. While OpenAI confirms the feature is powered by the GPT-5.6 family, the underlying token limits and context capacities for the Sol, Terra, and Luna models are not described in public materials.

Where it fits

The operator read is that ChatGPT Work bridges the gap between raw model intelligence and applied task orchestration. By combining the reasoning capabilities of GPT-5.6 with Codex, it fits into workflows that require heavy, multi-app synthesis—such as pulling a product spec from a connected drive, cross-referencing it with an explicitly linked Slack thread, and generating a formatted presentation.

The emerging pattern suggests a shift toward outcome-based delegation rather than simple prompt-and-response. Because users on premium tiers can set an effort level and leave the agent to run for hours, operators should view this as a potential asynchronous digital worker for document and code generation, provided the required apps are securely and intentionally integrated.