OpenAI has launched ChatGPT Work, an AI agent designed to autonomously execute complex workplace tasks for hours at a time.
Unveiled alongside the broader release of GPT-5.6, ChatGPT Work combines OpenAI's popular chatbot with its Codex coding tool to create documents, spreadsheets, presentations and websites.
The agent connects with Microsoft 365, Google Drive, Slack and Notion, and ships with enterprise governance controls, real-time monitoring and automated red-team security evaluations.
The launch marks OpenAI's most direct play yet for the enterprise market, as the IPO-bound company battles Anthropic – which launched its own autonomous agent, Claude Cowork, in January – for the lucrative business contracts that dwarf consumer subscription revenues.
A Model Built For The Enterprise?
GPT-5.6 launches in three tiers: Sol, the flagship model for complex reasoning; Terra, aimed at mainstream enterprise applications; and Luna, designed for high-volume, lower-cost deployments.
Sol is priced at $5 per million input tokens and $30 per million output tokens, and OpenAI claims it is 54 percent more token-efficient on agentic coding tasks than rival models.
Speaking to CNBC, OpenAI CEO Sam Altman explained: “If you want broad access, which we do, and you have powerful models, you really want to be able to be confident in your safety claims, because otherwise the world is going to get uncomfortable very fast.”
The company is pitching GPT-5.6 as competitive with far more expensive models – at twice the speed and significantly lower cost.
For CIOs who have grown wary of eye-watering AI infrastructure bills, the pricing story matters. But cost efficiency alone won't be enough to win over enterprise buyers staring down the prospect of an AI agent operating autonomously inside their systems for hours at a time.
The Trust Gap
That is arguably the harder sell. ChatGPT Work is designed to translate broad user goals into completed work with minimal human input – gathering context from connected apps, executing multi-step tasks and producing finished deliverables.
It is a compelling pitch. It is also one that asks organisations to extend a significant degree of operational trust to a system they cannot fully observe in real time.
OpenAI has attempted to address this head-on, building governance and security features directly into the product, including real-time monitoring and automated red-team evaluations designed to stress-test the agent before deployment.
