OpenAI Launches ChatGPT Work: The Super App That Turns AI Into an Employee

OpenAI’s long-anticipated enterprise offensive arrived this week with the launch of ChatGPT Work, a platform that fundamentally redefines what the company’s flagship product can do. Announced Thursday, July 9, and now rolling out to Pro and Enterprise users, ChatGPT Work is less a chatbot upgrade and more a wholesale reconstruction of how AI integrates into professional workflows — one that OpenAI hopes will finally displace Anthropic’s growing grip on the enterprise segment.

What ChatGPT Work Actually Does

At its core, ChatGPT Work is an autonomous execution environment. Users hand it an outcome — a quarterly report, a competitive analysis, a deployable web app — and the agent takes over. It breaks the goal into smaller steps, gathers information across connected apps and workflows, and stays with the project for hours if necessary, operating independently and surfacing finished materials rather than drafts requiring human revision.

The output is concrete: spreadsheets, slide decks, polished documents, and functional web apps, all generated without the back-and-forth that has characterized AI-assisted work to date. OpenAI officials described it as the difference between a tool that gives you ideas and one that “ships finished work.”

The platform runs on GPT-5.6, the company’s newest flagship model family introduced alongside ChatGPT Work. The Sol variant of GPT-5.6 — optimized for agentic coding and computer-use tasks — posted 88.8% accuracy on Terminal-Bench 2.1, a developer-terminal coding benchmark, narrowly edging Anthropic’s Fable 5 at 88%. Altman told CNBC that Sol is 54% more token-efficient on agentic coding tasks than its predecessor, a figure that translates directly to cost savings for enterprise deployments running agents at scale.

Merging the Product Family

The ChatGPT Work launch also reorganizes OpenAI’s increasingly sprawling product lineup. The standalone Codex application — which OpenAI introduced earlier this year as its coding agent — is folding into a redesigned ChatGPT desktop application that puts Chat, Work, and Codex under a unified interface. Notably, the new desktop app makes all three available on the Free tier, not just paid plans — a significant broadening of access.

Workspace agents represent another new layer. Enterprise and Business users can now create custom agents, test them, share them within their organizations, and schedule them to run automatically, including as integrations inside Slack for Business and Enterprise workspaces. The agents can connect to internal data sources, giving them the operational context required to function within a specific company’s environment rather than starting from scratch on every session.

Targeted Directly at Claude Cowork

The competitive framing inside OpenAI’s announcement was unusually direct. Multiple executives described ChatGPT Work as a direct response to Anthropic’s Claude Cowork, the agentic workspace product Anthropic launched in January. Cowork has accumulated a substantial enterprise customer base over the past six months, particularly at financial services, legal, and professional services firms where complex, multi-step document workflows are a daily requirement.

Anthropic had the head start, and it mattered. That head start gave Anthropic considerable leverage in enterprise procurement conversations, as IT buyers who built internal processes around Cowork face nontrivial switching costs. OpenAI officials said ChatGPT Work would be “both cheaper and more broadly available” than rival offerings — a pointed jab at both Cowork’s enterprise pricing tier and at xAI’s Grok 4.5, which debuted last week at a premium price point.

The pricing structure for ChatGPT Work has not been fully disclosed, but OpenAI indicated it would be bundled into existing ChatGPT subscriptions rather than sold as a standalone product — effectively offering agentic capabilities at a lower incremental cost to users already paying for Pro or Business access.

A Broader Platform Consolidation

ChatGPT Work is part of a larger consolidation happening across the AI assistant landscape. As frontier model capabilities converge — GPT-5.6 Sol, Claude Fable 5, and Grok 4.5 are now separated by single-digit percentage points on most standard benchmarks — the competitive battleground has shifted to workflow integration, agent reliability at scale, and enterprise distribution.

OpenAI’s bet is that bundling agentic capabilities into the world’s most recognized AI product will overcome Anthropic’s first-mover advantage. Anthropic’s counter-bet is that enterprise IT teams, once deeply integrated with Cowork’s agent architecture, will prove sticky. The next two quarters of enterprise deal flow will determine which thesis wins.

Both companies are also reportedly preparing for potential initial public offerings, making enterprise revenue figures increasingly scrutinized by institutional investors. ChatGPT Work gives OpenAI a direct answer to the question of how it plans to monetize AI at the next scale — not by licensing access to a smart chatbot, but by billing for autonomous, completed work.

What Developers Should Watch

For developers building on the OpenAI platform, the Work rollout introduces workspace agent APIs that streamline much of what previously required complex orchestration through the Assistants API. The new endpoints simplify agent creation, scheduling, and enterprise environment integration.

Codex’s absorption into the unified ChatGPT desktop app also signals a philosophical shift: OpenAI now treats code generation as one capability within a general-purpose work platform, not a specialized standalone tool. Developers who built workflows around the standalone Codex app should monitor the transition timeline, as OpenAI plans to phase out the standalone version in the coming weeks.

The broader message from this week’s launch is unmistakable: OpenAI sees the enterprise as the central battleground for the next phase of AI adoption, and it is no longer content to cede that ground to Anthropic while competing primarily on model benchmarks.