ChatGPT Work Is Here: How OpenAI Merged ChatGPT and Codex Into a Single Autonomous Agent -

Key Takeaways:

OpenAI launched GPT-5.6 in three tiers (Sol, Terra, Luna) and a new product called ChatGPT Work that merges ChatGPT and Codex into one autonomous agent.
ChatGPT Work can execute multi-step workflows across apps like Slack, Google Drive, Salesforce, and GitHub — staying with projects for hours.
GPT-5.6 Sol claims an 80-point score on the Artificial Analysis Coding Agent Index, surpassing Anthropic’s Fable 5 while using fewer tokens and lower cost.
Pricing ranges from $1/$6 per million tokens (Luna) to $5/$30 per million tokens (Sol), with usage-based billing for agent tasks.
The launch comes two weeks after a Trump administration limited preview period and coincides with COO Fidji Simo’s departure.

On July 9, 2026, OpenAI did something it had been building toward for two years: it killed the boundary between chatting and working. The company launched GPT-5.6 — its latest frontier model family — alongside a new product called ChatGPT Work that fuses the conversational interface of ChatGPT with the autonomous execution capabilities of Codex into a single product.

Sam Altman called GPT-5.6 “the best model we have ever produced.” But the real story isn’t the model. It’s what OpenAI built on top of it.

What ChatGPT Work Actually Does

ChatGPT Work is an agent mode inside ChatGPT that can operate across your connected apps, files, and workflows — autonomously, for hours at a time. Give it a goal like “analyze this quarter’s sales data and prepare a board-ready summary,” and it pulls data from your CRM, cross-references it with financial documents in Google Drive, builds charts, writes the narrative, and delivers a finished artifact.

The product introduces a Unified Plugins Directory that bundles third-party integrations in one place. At launch, the directory includes Google Drive, SharePoint, Slack, Microsoft Teams, Gmail, Outlook, Salesforce, Adobe, Zoom, LinkedIn, GitHub, Canva, Dropbox, and calendars. Users invoke specific plugins with an “@” mention — type “@Salesforce” to pull pipeline data, or “@Slack” to post an update — or let the agent figure out which tools it needs.

This echoes OpenAI’s failed plugin push from 2023. OpenAI co-founder Greg Brockman recently admitted those plugins “didn’t work at all because the models weren’t ready.” With GPT-5.6 behind it, the company expects a different outcome.

The desktop app is the most capable surface. ChatGPT Work can access and modify local files, use a built-in browser to navigate web tools, and run code through integrated Codex capabilities. OpenAI also introduced a feature called Sites in public beta, which lets users transform work materials into shareable interactive websites — dashboards, project trackers, prototypes, and reports that auto-update when source data changes.

The GPT-5.6 Model Family: Sol, Terra, and Luna

GPT-5.6 ships in three variants, each targeting a different cost-performance balance:

Sol ($5 input / $30 output per million tokens) — the flagship, designed for advanced reasoning, coding, and complex knowledge work.
Terra ($2.50 / $15) — a balanced model for everyday enterprise workloads, competitive with GPT-5.5.
Luna ($1 / $6) — the fastest and most affordable, optimized for high-volume, low-complexity tasks.

On the Artificial Analysis Coding Agent Index, GPT-5.6 Sol scores 80 — 2.8 points above Anthropic’s Fable 5 — while using less than half the output tokens and costing roughly one-third less. The model also achieves a 54% improvement in token efficiency over GPT-5.5 for agentic coding tasks, according to Altman’s CNBC interview.

A new “ultra” reasoning mode coordinates four parallel agents to tackle demanding tasks faster, reaching higher scores on BrowseComp (92.2%), SEC-Bench Pro, and Terminal-Bench 2.1. For developers building multi-agent systems, OpenAI added a multi-agent beta in the Responses API.

One notable technical addition is Programmatic Tool Calling in the Responses API — GPT-5.6 can write and run lightweight programs that coordinate tools, process intermediate results, and adapt workflows without routing every interaction back through the model. This makes it Zero Data Retention compatible.

The Regulatory Backstory

The public launch came two weeks after a turbulent limited preview period. In late June 2026, the Trump administration restricted GPT-5.6 access to select government-approved organizations during what OpenAI called a “limited preview.” The company received the greenlight for public rollout on July 9.

The same day, OpenAI announced that COO Fidji Simo — who joined just 11 months earlier — was stepping down due to worsening health issues. Simo had been leading OpenAI’s AGI efforts and consumer product strategy.

How ChatGPT Work Compares to Claude Cowork

ChatGPT Work is a direct competitor to Anthropic’s Claude Cowork, which Anthropic launched for web and mobile on July 7 — just two days before OpenAI’s announcement. Both products combine a coding agent with a general-purpose AI assistant, both operate across connected tools, and both can work autonomously on complex tasks for extended periods.

The key differences come down to execution. ChatGPT Work is available on desktop (Windows and Mac) for all users, including free tier, immediately. On web and mobile, it rolls out first to Pro, Enterprise, and Edu subscribers. Claude Cowork’s web and mobile beta is rolling out over several weeks, starting with Max users.

Pricing structures also differ. ChatGPT Work shares the same consumption pool as Codex, with usage-based billing that scales with task complexity and model selection. Claude Cowork operates on Anthropic’s subscription tiers with doubled usage limits through August 5.

The old ChatGPT desktop app has been rebranded “ChatGPT Classic” — a deprecated version accessible only through a “quick chat” button. The existing Codex app now redirects to the unified ChatGPT desktop app.

Enterprise Controls and Security

OpenAI is betting that enterprises will adopt ChatGPT Work only if they can control what it touches. The product includes an Auto-Review feature where advanced models check important actions before execution. OpenAI claims Auto-Review blocked 100% of attempts to extract protected data during adversarial red-teaming.

Enterprise and Edu administrators can set spend controls at the workspace, group, and individual user levels. The Compliance API lets organizations pre-authorize which actions ChatGPT Work is allowed to take, which plugins it can access, and how much it can spend.

For IT teams worried about an AI agent running loose across company systems, OpenAI positions these controls as a key differentiator — though competitors offer similar governance features.

What This Means for Knowledge Workers

The ChatGPT Work launch signals a shift in how AI companies think about their products. ChatGPT started as a conversational tool. Codex was for developers. Now they’re the same thing, aimed at anyone who works with information.

The practical impact: a marketing manager can ask ChatGPT Work to pull last quarter’s campaign data from Salesforce, analyze performance against goals in a spreadsheet, draft a strategy memo, and create a presentation — all from a single prompt, without switching between five different tools. A finance analyst can have it reconcile budget data across multiple systems and produce a variance report ready for review.

The catch is cost. ChatGPT Work tasks consume from the same credit pool as Codex, and complex agent workflows can burn through tokens quickly. OpenAI warns that Work tasks “may use more of your plan’s included usage.” Free and Go users get Terra only; Sol and ultra reasoning modes require Pro or Enterprise plans.

The Bigger Picture

ChatGPT Work arrives in a week of AI agent announcements. Anthropic launched Claude Cowork on web and mobile. Grok 4.5 debuted from SpaceXAI on July 8 with strong coding capabilities at aggressive pricing ($2/$6 per million tokens). Google made AlphaEvolve generally available for algorithm optimization on Cloud.

The pattern is clear: every major AI lab is converging on the same product — an autonomous agent that connects to your tools, reads your data, and produces finished work. The question is no longer whether AI agents will transform knowledge work. It’s which agent you’ll trust with your systems, your data, and your budget.

Frequently Asked Questions

Is ChatGPT Work available now?
Yes. It’s available immediately on the ChatGPT desktop app for Windows and Mac across all plans, including free tier. On web and mobile, Pro, Enterprise, and Edu users get access first, with Plus and Business following in the coming days.

How much does ChatGPT Work cost?
ChatGPT Work uses the same consumption pool as Codex. Pricing depends on the model tier (Sol, Terra, or Luna) and task complexity. Sol costs $5 input / $30 output per million tokens; Terra costs $2.50 / $15; Luna costs $1 / $6.

Can ChatGPT Work access my local files?
Yes, through the desktop app. The web and mobile versions work through connected cloud services like Google Drive and SharePoint.

What is the difference between ChatGPT Work and Codex?
Codex technology is built into ChatGPT Work. Codex remains available as a separate “view” inside the ChatGPT app alongside the “Work” view. The distinction: Codex focuses on code-related tasks; ChatGPT Work extends to all knowledge work.

Is ChatGPT Work safe for enterprise use?
OpenAI includes Auto-Review, spend controls, and Compliance API tools. Admins can restrict which actions the agent can take and which tools it can access. During adversarial red-teaming, Auto-Review blocked 100% of data extraction attempts, according to OpenAI.

References

OpenAI. “GPT-5.6: Frontier intelligence that scales with your ambition.” openai.com/index/gpt-5-6. Published July 9, 2026.
OpenAI. “ChatGPT is now a partner for your most ambitious work.” openai.com/index/chatgpt-for-your-most-ambitious-work. Published July 9, 2026.
Field, Hayden. “OpenAI rolls out GPT-5.6 after government greenlight — and announces ‘ChatGPT Work’.” The Verge. Published July 9, 2026.
Bastian, Matthias. “OpenAI pairs its GPT-5.6 public rollout with ChatGPT Work.” The Decoder. Published July 9, 2026.
Orland, Kyle. “OpenAI wants its new tool to do your work for you and with you.” Ars Technica. Published July 9, 2026.
Wheatley, Mike. “OpenAI debuts ChatGPT Work, an agentic tool for automating business workflows.” SiliconANGLE. Published July 9, 2026.
Clover, Juli. “OpenAI Debuts ChatGPT Work Agent and New GPT-5.6 Models.” MacRumors. Published July 9, 2026.