OpenAI expands ChatGPT Work to automate office tasks

OpenAI is actively shifting its focus from software developers to general office workers with ChatGPT Work, a $20-per-month subscription tier designed to run autonomous digital agents. Built on the foundation of the company's Codex coding tool, the product aims to help non-engineers automate multi-step tasks across everyday applications. While software developers have rapidly adopted agentic tools, OpenAI is trying to bridge a massive adoption gap. An internal study from June revealed that 98 percent of OpenAI employees were using Codex, compared to just 17 percent of organizational subscribers and less than 1 percent of individual subscribers.

The joint app for Codex and ChatGPT Work currently has 20 million users, a small fraction of the more than 1 billion people who prompt ChatGPT online. To make the tool accessible to non-technical professionals, OpenAI engineers are designing a more intuitive user interface, contrasting with command-line tools like Claude Code from rival Anthropic. While Anthropic's Claude Cowork relies on a conversational back-and-forth style, OpenAI is betting on the raw reasoning power of its latest models. To evaluate these agents across 44 occupations, OpenAI uses its GDPval benchmark alongside direct user feedback.

However, running these complex agents is highly resource-intensive. During a four-day trial, a TechCrunch reporter consumed over 80 million tokens, racking up an estimated cost of $65, which represents a subsidy of more than 3x the monthly subscription price. To address these high operational costs, OpenAI is focusing on efficiency, recently implementing an 80 percent price cut for its Luna model. Additionally, the company faces competition from open-source alternatives. For instance, the minimalist Pi harness developed by Earendil's Mario Zechner, which powers projects like OpenClaw and Cloudflare OS, outperformed Codex on benchmarks while running on the same GPT 5.5 model.