ChatGPT Work Launches: OpenAI Enters the Agentic Workspace Race Against Claude Cowork and Grok Build

OpenAI launched ChatGPT Work on July 9, 2026, and with it, the agentic workspace market officially became a four-way fight. ChatGPT Work is an agent-based product that completes full multi-step workplace tasks autonomously — not conversational assistance, but autonomous execution from a single instruction. It runs on GPT-5.6 and Codex, and it represents OpenAI's most direct attempt yet to capture enterprise workflows.

What ChatGPT Work Actually Does

ChatGPT Work is an AI agent that lives in the cloud and acts across your apps and files. It connects to Slack, Microsoft Teams, Google Drive, SharePoint, GitHub, email, and calendars. Give it a task like "prepare the Q3 budget review with actuals from our finance sheet, pull the latest revenue data from Salesforce, draft the presentation, and email it to the leadership team" — and it executes the whole chain.

OpenAI merged its Codex and ChatGPT desktop applications into a single unified app for Mac and Windows. The new app has three modes:

• <<<BOLD>>>Chat<<<BOLDEND>>> — standard conversational AI, file uploads, web search, image generation
• <<<BOLD>>>Codex<<<BOLDEND>>> — the AI coding environment, now embedded in the same app
• <<<BOLD>>>Work<<<BOLDEND>>> — the autonomous agent mode that can stay with a project for hours

The unification matters. Users don't need to switch between a coding tool and a chat tool. The same agent can write code, draft documents, send messages, and browse the web — all in one session. OpenAI is betting that the all-in-one workspace will be stickier than point solutions.

The Four-Way Agentic Workspace Market

The competitive landscape solidified into four distinct products this week:

<<<BOLD>>>ChatGPT Work<<<BOLDEND>>> — $200/month (ChatGPT Pro tier). Powered by GPT-5.6 Sol. Best integration with Microsoft 365 ecosystem. Three-mode unified app. Weakness: limited third-party tool integrations compared to Anthropic's MCP ecosystem.

<<<BOLD>>>Claude Cowork (Anthropic)<<<BOLDEND>>> — $200/month (Claude Max tier). Powered by Claude Sonnet 5 and Claude Opus 4. 3,000+ MCP integrations. Scores 81.2% on OSWorld computer use benchmark — the highest publicly reported result. Strengths: tool ecosystem, computer control, document analysis. Weakness: no unified desktop app (still browser-based).

<<<BOLD>>>Muse Spark 1.1 (Meta)<<<BOLDEND>>> — API pricing $1.25/input, $4.25/output per million tokens. Launched July 9 as Meta's first paid model. 1 million context window. Supports MCP protocol. Built for agentic and coding tasks. Weakness: no independent SWE-bench Pro score yet, and Meta has no dedicated agentic workspace product — it's selling the model, not the experience.

<<<BOLD>>>Grok Build (SpaceXAI)<<<BOLDEND>>> — API pricing $3/input, $6/output. Live X data integration. Real-time social data feeds give it an edge for marketing and sentiment analysis tasks. Weakness: highest pricing in the tier, and the Elon Musk factor makes some enterprise buyers cautious.

Pricing and Benchmark Comparison

The pricing tiers align closely. ChatGPT Work and Claude Cowork both cost $200/month for the top tier. The model-level pricing tells a different story:

• GPT-5.6 Sol: $5/input, $30/output per million tokens
• Claude Sonnet 5: $4/input, $26/output
• Muse Spark 1.1: $1.25/input, $4.25/output
• Grok 4.5: $3/input, $6/output

Meta is undercutting everyone on price by a wide margin — roughly 6x cheaper than OpenAI on output tokens. But Meta lacks the integrated workspace product. It's selling the engine, not the car.

On benchmarks, Claude Cowork leads on OSWorld (computer control). GPT-5.6 Sol leads on SWE-bench Verified (software engineering). Muse Spark 1.1 has no independent benchmark scores yet. Grok 4.5 leads on real-time data integration but trails on general reasoning.

What the Enterprise Buyer Sees

For companies choosing an agentic workspace provider, the decision comes down to three factors:

<<<BOLD>>>Tools and integrations.<<<BOLDEND>>> Anthropic's MCP ecosystem is the deepest. 3,000+ connectors means Claude Cowork can already talk to most enterprise tools. OpenAI's plugin ecosystem is growing but smaller. Meta's model has MCP support but no product to speak of yet.

<<<BOLD>>>Data privacy.<<<BOLDEND>>> ChatGPT Work runs in OpenAI's cloud. Claude Cowork runs in Anthropic's cloud. Enterprises processing sensitive data need to trust the provider. Microsoft 365 customers will find ChatGPT Work's native integration attractive. Regulated industries may prefer Anthropic's documented privacy stance.

<<<BOLD>>>Ecosystem lock-in.<<<BOLDEND>>> OpenAI's tight integration with Microsoft 365 is the strongest ecosystem play. If your company lives in Teams, SharePoint, and Outlook, ChatGPT Work is the natural choice. Claude Cowork works everywhere but has no native integration with any major productivity suite.

What to Watch

The next 30 days will be decisive. Meta needs to publish independent SWE-bench Pro scores for Muse Spark 1.1 — without them, developers won't trust the quality claims. OpenAI needs to grow its MCP connector count to match Anthropic's depth. Anthropic needs to ship a unified desktop app — browser-only is a weakness for the "always-on" use case.

The agentic workspace market is projected to reach $47 billion by 2028. The winner won't be the company with the best model. It will be the company that delivers the most reliable autonomous task execution across the most enterprise tools. That's a product and integration challenge, not an AI research problem.

FAQ

Q: Is ChatGPT Work replacing ChatGPT Agent?

A: Yes. OpenAI consolidated Operator, deep research, and ChatGPT Agent into ChatGPT Work. It's the new flagship agent product for both individual and enterprise users.

Q: Can I use ChatGPT Work without ChatGPT Pro?

A: No. ChatGPT Work requires the $200/month ChatGPT Pro subscription. OpenAI sees agentic features as the top-tier value proposition.

Q: How does ChatGPT Work compare to Claude Cowork on actual task completion?

A: Published benchmarks are limited. OpenAI claims ChatGPT Work completes multi-step tasks with 87% success rate in internal testing. Anthropic hasn't published comparable numbers for Claude Cowork's task completion rate.

Q: Does ChatGPT Work support custom APIs and integrations?

A: Yes, through OpenAI's Actions framework and the GPT Store. Users can build custom connectors, but the available library is smaller than Anthropic's MCP marketplace.