Enterprise

OpenAI launches ChatGPT Work, an agent aimed at hours-long enterprise tasks

Powered by the new GPT-5.6 model family, ChatGPT Work turns the chatbot into an autonomous workplace agent — landing squarely opposite Anthropic's Claude Cowork and Microsoft's Copilot Cowork.

By Asher Glick · Enterprise software desk · 2026-07-20 · 3 min read

Photo: Unsplash / Andrew Neel — A person working at a laptop on a wooden desk in a modern office setting

OpenAI shipped ChatGPT Work on July 9, 2026, alongside a new GPT-5.6 model family (Sol, Terra, Luna), reframing its consumer chatbot as an autonomous agent for what the company describes as “longer, more involved tasks.” An “ultra” setting scales the system up to 16 parallel agents. Codex has been folded into a redesigned desktop app, and Scheduled Tasks let the agent run without a user in the loop.

The competitive geometry is the actual story. ChatGPT Work follows Anthropic’s Claude Cowork, which can plan and execute multi-step work across legal, sales, marketing, and data-analysis plug-ins, and whose debut triggered a sharp selloff in U.S. and European software and professional-services stocks. Microsoft’s Copilot Cowork sits on the other flank. OpenAI, citing the Artificial Analysis Coding Agent Index, claims GPT-5.6 outshines Anthropic’s models “at every turn,” and touts Sol as 54% more token efficient on agentic coding tasks.

The wrinkle is Microsoft. It’s an OpenAI backer, a Copilot Cowork vendor, and, as of July 14, has confirmed GPT-5.6 as the preferred model inside Microsoft 365 Copilot. Partner and competitor on the same surface.

Rollout is tiered. Pro, Enterprise and Edu users got web and mobile access first, with Plus and Business following over subsequent days. Enterprise and Edu workspaces received a two-week preview with Work off by default, and are being auto-activated the week of July 20–23. GPT-5.6 itself was held back at the request of the U.S. government, limited to a “small group of trusted partners” before broader release roughly two weeks later.

Sam Altman offered the framing that matters to the sales motion: “every enterprise now is thinking about spend and the value they’re getting in exchange for AI.” Bloomberg read the launch as OpenAI’s response to Anthropic and Microsoft, and as positioning ahead of an anticipated IPO. Both readings are correct, and neither is flattering. The frontier lab is now selling agent-hours to CFOs who spent 2025 auditing last year’s Copilot invoices, and it’s doing so against a partner whose distribution it depends on.

Sources