Understanding ChatGPT Work
ChatGPT Work is split into two products: Work Cloud (accessible via chatgpt.com and mobile apps) and Work Local (desktop app with file/program access), with the article focusing on Work Cloud Work Cloud offers unique capabilities absent in regular Chat: code execution with full internet access, a headless Chrome browser with JavaScript execution, persistent filesystem across sessions, sub-agent delegation, and model selection (GPT-5.6 Sol/Luna/Terra with reasoning levels) The code execution envi
Analysis
TL;DR
- ChatGPT Work is split into two products: Work Cloud (accessible via chatgpt.com and mobile apps) and Work Local (desktop app with file/program access), with the article focusing on Work Cloud
- Work Cloud offers unique capabilities absent in regular Chat: code execution with full internet access, a headless Chrome browser with JavaScript execution, persistent filesystem across sessions, sub-agent delegation, and model selection (GPT-5.6 Sol/Luna/Terra with reasoning levels)
- The code execution environment's internet access is a major differentiator, allowing cloning of GitHub repos, package installation, and unrestricted web interaction—far surpassing Claude's restricted allowlist approach
- Work is gated behind $20/month subscriptions and billed against a separate Codex allowance, distinct from Chat's allowance, which explains model availability differences
- The persistent /workspace/scratch filesystem is shared across all running Work sessions, enabling cross-session file access but not shared process space or localhost server access
Why It Matters
ChatGPT Work represents OpenAI's strategic push to position ChatGPT as a task-completion engine rather than just a conversational interface, directly competing with Claude's coding and browser automation capabilities. The introduction of unrestricted internet-accessible code execution in a cloud environment is a significant capability leap that closes a gap many AI practitioners had noted since the original Code Interpreter launch. Understanding the Work vs. Chat distinction and the underlying billing/model architecture is essential for professionals evaluating which OpenAI product tier delivers the most value for their workflows.
Technical Details
- Architecture split: Work Cloud runs entirely in OpenAI's infrastructure (accessible via web/mobile), while Work Local runs on the user's machine via the desktop app (formerly Codex), with file and program access
- Model selection: Work offers GPT-5.6 Sol, Luna, and Terra with reasoning levels (Light/Medium/High/Extra High/Max/Ultra), plus GPT-5.5 at Light/Medium/High/Extra High; Chat offers a separate lineup including 5.6 Instant and 5.6 Pro (exclusive to Chat, $100+/month tier)
- Code execution with internet: The sandboxed code environment supports full internet access by default (configurable domain allowlists), enabling GitHub repo cloning, PyPI/NPM package installation, and arbitrary web API interaction—unlike Claude's short allowlist
- Headless Chrome browser: Built on Playwright, supports form filling, screenshots, JavaScript execution against the DOM, and secure credential handling (user enters passwords/2FA directly without model exposure)
- Persistent filesystem: Each Work session gets a /workspace/scratch/
folder that persists across sessions and is mounted across all running Work sessions simultaneously, though process space and localhost servers remain isolated per session - Sub-agent delegation: Ultra reasoning mode eagerly delegates to sub-agents; Work supports running sub-agent sessions with Sol, Luna, and Terra models
- Access control: Available only to $20/month subscribers and above; free and $8/month Go users are excluded
Industry Insight
- OpenAI is clearly segmenting its product stack along a "chat for information, work for execution" paradigm—practitioners should evaluate whether their workflows justify the $20/month upgrade based on the unique capabilities (internet-connected code execution, persistent storage, browser automation) rather than treating Work as merely a rebranded Codex
- The unrestricted internet access in the code execution sandbox represents a significant capability advantage over competitors like Claude, but also raises security and abuse considerations that may prompt OpenAI to introduce tighter default restrictions in future iterations
- The separate billing allowance for Work vs. Chat suggests OpenAI is experimenting with differentiated cost structures for high-compute tasks (sub-agent delegation, browser sessions, code execution)—professionals should monitor how these allowances evolve as the product matures and consider the total cost of ownership for heavy Work usage
Disclaimer: The above content is generated by AI and is for reference only.