Peter Steinberger, creator of OpenClaw and OpenAI employee, has revealed that his project consumed $1.3 million in API tokens in a single month. Using GPT-5.5 as the engine, they processed 603 billion tokens across 7.6 million requests, all covered by OpenAI as a research investment. A team of three orchestrates 100 Codex agents in parallel.
100 Codex agents automate code review and bug detection 🤖
The system deploys about 100 Codex agents in parallel for tasks such as code review, vulnerability detection, issue deduplication, automatic solution writing, and benchmark monitoring. Steinberger detailed that this architecture allows an open-source project to maintain the cadence of a much larger team. Requests are distributed among specialized agents that work without constant human intervention, optimizing each phase of development.
OpenAI's AI resolves issues while humans drink coffee ☕
While the human team takes a break, the 100 Codex agents handle code review, detect vulnerabilities, and even write solutions. The curious part is that the $1.3 million bill is paid by the same company that sells the tokens. It seems the strategy is to spend money on their own API to prove it works, even though the monthly budget of some indie projects is less than an intern's salary.