Peter Steinberger, the creator of OpenClaw and an OpenAI engineer, used approximately $1.3 million worth of OpenAI tokens within a 30-day period ending in mid-May 2026 [1, 2, 3]. He publicly shared a screenshot on May 15 showing this token usage on X, highlighting the scale of consumption [2].
The spending covered over 600 billion tokens and around 7.6 million OpenAI API requests during that period [1, 3]. Steinberger operated about 100 Codex-based AI agents managed by a small team of three people [1, 3]. These AI systems performed a variety of software development tasks including writing code, scanning for vulnerabilities, fixing bugs, reviewing pull requests, filtering GitHub issues, and even attending team meetings to generate code in real time [1, 3].
The $1.3 million bill was fully covered by OpenAI since Steinberger is an employee, confirming that the costs did not come out of his own pocket [1, 2, 3]. The pricing reflected Codex's "Fast Mode," which prioritizes speed but costs more; disabling Fast Mode could cut the monthly costs by about 70% to roughly $300,000 [1, 3].
Steinberger said the project is an experiment to explore "how would we build software in the future if tokens don't matter?" [2]. He reiterated the sentiment in a statement in Chinese expressing the goal of understanding what software development might look like if token costs were no longer a limitation [3].
OpenClaw, Steinberger's open-source AI product, has reportedly become viral and is noted as the fastest-growing AI tool of its kind. It has contributed to increased demand for hardware such as the Mac Mini [2, 3]. Steinberger also remarked that millions of people have enjoyed using OpenClaw, and that he is currently building a few startups in parallel [1].
The public sharing of the token usage offers rare transparency into the scale at which AI-driven software development can operate under minimal cost constraints. The experiment officially gained attention when Steinberger posted details on X on May 15, 2026 [2].
