AI sure uses a lot of resources. And if you're running a fleet of AI agents processing millions of requests, you'd probably expect the token bill to be pretty high. In the case of OpenClaw developer Peter Steinberger, that bill appears to equate to $1,305,088.81 for his OpenAI spend—all racked up in just 30 days.
The info was spotted in a screenshot Steinberger posted of his OpenAI dashboard. His account shows the aforementioned figure, with the top model being used attributed to GPT-5.5 (via Tom's Hardware). The bill represents 603 billion tokens across 7.6 million requests, all handled by around 100 Codex instances overseen by a team of three people.
Steinberger joined OpenAI in February, so according to him, it's the Sam Altman-led company that covers the costs. The Codex agents are said to be used for a number of tasks, including scanning for security vulnerabilities, writing fixes, and potentially even attending meetings as part of the continued development of OpenClaw, an open source autonomous AI agent.
Well, it is the "AI that actually does things." And developing all of those "things" looks to be pretty intensive on the token front—although this might be the tip of the iceberg. One X commenter asked, "bruh, is this your usage?", to which Steinberger responded, "Yup. At least on this account."
Others pointed out the bill would equate to the bankroll of a small startup, and to that Steinberger replied: "I'm building a few startups in parallel." Another cheekily commented, "$1.3m/month. Anything useful yet?" which earned a response of: "Other than millions of people enjoying OpenClaw? Yeah."
Ouch, the salt. Anyway, Steinberger also attributes the figure to the fact that it's representative of Codex's "Fast Mode" pricing, and that the API costs would be 70% cheaper if used traditionally.
Working for OpenAI has its perks, I guess. And while the AI industry hoovers up huge amounts of hardware for its compute needs, OpenClaw seems to be taking its share. Alright for some, isn't it?
