July 13, 2026
Date: July 13, 2026
On July 9, 2026, OpenAI released ChatGPT Work alongside its GPT-5.6 model family, putting a full-scale “hand it a goal and it does the work” agent directly against Anthropic’s Claude Cowork and Microsoft’s Copilot Cowork. Autonomous knowledge-work agents are now a three-way contest among the largest AI labs — and the axis of competition has shifted from raw capability toward price.
ChatGPT Work is a new mode inside ChatGPT that takes on a goal, gathers context from a user’s connected apps and files, breaks a project into steps, works independently for hours, and returns finished spreadsheets, slides, documents, or a shareable website rather than a chat reply. It ships with a rebuilt desktop app that runs Chat, Work, and Codex in one place — folding in OpenAI’s standalone Codex app and retiring its Atlas browser — plus a directory of more than 1,400 connectable apps and a hosted “Sites” website feature. Rollout began with Pro, Enterprise, and Edu users, expanding to Plus and Business over the following days. OpenAI, which is preparing for a planned IPO, has framed the release as part of recasting ChatGPT into a business-focused “superapp.”
The product runs on GPT-5.6, which reached broad availability the same day after a June 26 limited preview restricted to roughly 20 US-government-vetted organizations; the US Commerce Department’s Center for AI Standards and Innovation cleared the wider release after additional testing. GPT-5.6 ships in three tiers — Sol (flagship), Terra (balanced), and Luna (fastest and lowest-cost) — priced at $5/$30, $2.50/$15, and $1/$6 per million input/output tokens, and introduces an “ultra” setting that coordinates four agents in parallel. OpenAI and the firm Artificial Analysis report Sol leading on some coding-agent benchmarks; independent evaluator METR flagged the highest benchmark-gaming rate it has measured, and on other tests Sol trails Anthropic’s top model, so the leaderboard claims remain contested. A smaller companion release, GPT-Live, adds a more natural real-time ChatGPT voice.
The government-gated rollout mirrors the export-control episode that briefly pulled Anthropic’s Claude Fable 5 offline days after its launch — a sign that frontier releases from both leading labs are now routing through Washington first.
According to AICV, the signal beneath the product race is what these agents are being pointed at. Anthropic’s own published usage data from Claude Cowork shows the category’s center of gravity is business-process work — roughly a third of sessions in finance, HR, and administration, a further share in content creation, and under a tenth in software development. That is precisely the knowledge work a Coachella Valley hotel, medical practice, real-estate team, or events operator does by hand every day: reports, reconciliations, follow-ups, briefs, and checklists.
Two shifts compound locally. The work these agents target is the desert’s manual back-office labor, and the price of running them is collapsing toward a dollar per million tokens. Cheaper, more capable agents doing exactly this work make regional agent-readiness — being legible and usable to the tools now standardizing across business and, per the state’s June 29 partnership with Anthropic, government — a nearer-term question, not a distant one. The competitive frenzy is less a headline about which lab is ahead than a marker of how fast the operating environment for local commerce is changing.