OpenAI API bill tops $1.3M

Peter Steinberger, the founder of OpenClaw and an OpenAI employee since February, is at the center of a new burst of attention after screenshots circulated showing more than $1.3 million in OpenAI API charges over 30 days. Let’s Data Science, citing Tom’s Hardware and The Decoder, reported Sunday that the screenshots showed $1,305,088.81 in usage tied to 603 billion tokens and 7.6 million requests. The reports said the spending was linked to roughly 100 Codex instances used by Steinberger’s small team for software-development work. The same reports said OpenAI, not Steinberger personally, was covering the bill. ### Where did the $1.3 million number come from? The $1,305,088.81 figure came from screenshots that Steinberger shared from a usage dashboard, according to The Decoder and the aggregated report published by Let’s Data Science. Those reports said the dashboard covered a 30-day period and listed 603 billion tokens and 7.6 million requests alongside the dollar total. (letsdatascience.com) May 15 was one of the daily data points visible in follow-on coverage. OfficeChai, citing the same screenshots, reported a single-day spend of $19,985.84, about 19 billion tokens and 206,000 requests, with GPT-5.5-2026-04-23 listed as the top model. ### Who is Peter Steinberger, and what is OpenClaw? (letsdatascience.com) Peter Steinberger wrote on February 14 that he was joining OpenAI “to work on bringing agents to everyone” and that OpenClaw would move to a foundation while staying “open and independent.” In that post, he said OpenAI had made commitments that would let him continue dedicating time to the project and that the company already sponsored it. (officechai.com) OpenClaw’s GitHub repository describes the software as a “personal AI assistant” that runs on a user’s own devices and connects to a long list of messaging and workplace channels. The repository’s contributing page names Steinberger as maintainer, and the project remained active on GitHub on May 17, with frequent commits and more than 370,000 stars visible in the repository snapshot returned by web search. (steipete.me) ### What was the system apparently doing with that usage? The Decoder reported that a team of about three people was keeping roughly 100 Codex instances running in the cloud. The report said those agents reviewed pull requests, scanned commits for security issues, deduplicated issues, wrote fixes, monitored benchmarks and opened pull requests, including work triggered by meeting discussions. (github.com) Let’s Data Science, again attributing the details to Tom’s Hardware and The Decoder, described the same workload as an automated developer pipeline spread across code review, vulnerability scanning, issue triage and benchmark monitoring. The reports also said GPT-5.5 was the top model on the dashboard. ### Was Steinberger actually paying that bill himself? (the-decoder.com) OpenAI was covering the cost, according to The Decoder and the reports cited by Let’s Data Science. Those reports framed the spending as part of a broader experiment in what software development looks like when token costs are not the main constraint. (letsdatascience.com) The Decoder attributed to Steinberger a defense of the setup as a research investment. The report also said he had argued that turning off “Fast Mode” would cut costs by 70% and that he viewed the return on investment as high because the team’s output was open source. ### What can readers verify next? (the-decoder.com) OpenClaw’s public GitHub repository, maintainer page and README remain live as of May 17 and show the project is still active and still presented as open source. Steinberger’s February 14 blog post remains the clearest primary-source statement on the project’s status inside OpenAI’s orbit: he said OpenClaw would move to a foundation and “stay open and independent.” (github.com) (the-decoder.com)