OpenAI Launches ChatGPT Work and Merges Codex Into Desktop — The Agentic Workspace Race Has Four Competitors

What ChatGPT Work Actually Is

OpenAI is launching ChatGPT Work, an agent-based product powered by Codex and the now publicly available GPT-5.6. The product is designed to complete full multi-step workplace tasks autonomously — not to answer individual questions, but to take on an entire work process and see it through. The positioning is explicit: Work is for tasks that previously required a human to stay in the loop at every step. Research a topic and draft a report. Find a time slot and send a meeting invite. Summarise 50 emails and extract action items. Process a dataset and generate a chart. All without being prompted after the initial instruction.

The Codex-and-ChatGPT desktop merge is the structural change that makes Work possible at scale. OpenAI merged Codex and ChatGPT desktop apps for Mac and Windows under a new ChatGPT desktop app, allowing users to switch between Codex, Chat, and Work. Previously, Codex was a separate tool for software engineering tasks. Chat was the conversational interface. Work is the new agentic layer. The three modes share context, memory, and file access from a single install — a unified desktop AI environment that more closely resembles an operating system integration than a chat application.

The Direct Competition — Claude Cowork and What Differs

ChatGPT Work is OpenAI's direct response to Claude Cowork — Anthropic's desktop agentic product launched in June 2026. The competitive positioning is clear: both products target the "complete a full work task for me" use case. The meaningful differences at launch: ChatGPT Work runs on GPT-5.6 (Terra for general tasks, Sol for code) with Codex integration in the same app. Claude Cowork runs on Claude Sonnet 5 and Fable 5 with 3,000+ MCP integrations and deep computer use via Claude's OSWorld-leading 81.2% computer use benchmark score.

The meaningful question for choosing between them: ChatGPT Work has deeper integration with the Microsoft 365 ecosystem (Word, Excel, Teams, Outlook) via existing Copilot relationships. Claude Cowork has broader third-party tool connectivity through MCP. For teams already in Microsoft 365, ChatGPT Work's ecosystem advantage is real. For teams using a diverse SaaS stack (Jira, Linear, Notion, Slack, GitHub), Claude Cowork's MCP depth likely wins. Neither product has been independently benchmarked on real workplace task completion. User testing will determine which framing holds up under production conditions.

Also: Meta Muse Spark 1.1 — First Paid Meta Model, Aims at Codex and Claude Code

Meta shipped Muse Spark 1.1 on July 9, 2026, its most capable model yet for real-world coding and agentic tasks, and started charging developers to use its own model for the first time through the new Meta Model API. Pricing: $1.25 per million input tokens and $4.25 per million output tokens — with $20 in free credits for new accounts. Context window: 1 million tokens. It supports MCP and custom-skill integration, and speaks both the OpenAI and Anthropic SDK formats, meaning switching to Muse Spark is a base-URL and API key change for most existing integrations. Thinking mode is free inside the Meta AI app. US-only at API launch.

The positioning is direct: Meta jumps into AI coding market in effort to chase Anthropic and OpenAI. At $4.25/M output, Muse Spark 1.1 undercuts Claude Sonnet 5 ($10/M intro), GPT-5.6 Terra ($15/M), and Claude Opus 4.8 ($25/M) on output price while offering 1M context and agentic task support. No independent SWE-bench Pro score has been published for Muse Spark 1.1 yet. Given that this is Meta's first paid model and first serious coding API offering, the benchmark validation will be the critical test of whether the price-performance story holds up. Watch for third-party evaluations over the next two weeks.

The Agentic Workspace Market — Four Products Now Competing

Sources: LLM Stats July 10-11, 2026 · FelloAI Best AI Models July 2026 · Related: GPT-5.6 Sol/Terra/Luna full review → · How to build profitable AI agents → · SuperGrok vs Claude Pro vs ChatGPT Plus →