Peter Steinberger’s 100 AI agents raised $1.3 million in OpenAI tokens in 30 days to build OpenClaw
The TL;DR
OpenClaw creator Peter Steinberger spent 1.3 million OpenAI API tokens in 30 days using 100 Codex instances in his open source project. This bill, put together by OpenAI where Steinberger currently works, represents 603 billion tokens across 7.6 million requests and provides a tangible point of public data at the expense of AI code independence at scale.
Peter Steinberger, founder of OpenClaw and developer at OpenAI, collected $1.3 million in API fees in one month by running nearly 100 Codex sessions at once on his open source project. This bill, which included 603 billion tokens across 7.6 million requests within 30 days, is the most visible demonstration yet of what happens when AI-powered software development is done without budget constraints, and how quickly costs rise when independent agents operate continuously at scale.
Steinberger posted a screenshot of the bill to X, showing $1,305,088.81 billed to the OpenAI API, with GPT-5.5 as the primary model. OpenAI covers expenses: Steinberger joined the company in February 2026, and the spending is considered a research investment in understanding what software development looks like when the token economy is not a limiting factor.
Peter Steinberger X Post – source: X
What agents actually do
The 100 Codex scenarios are not limited to codex generation. Steinberger’s three-person team built an independent development pipeline in which AI agents perform a series of tasks that would require a much larger engineering organization. Agents review pull requests, scan commits for security vulnerabilities, issue duplicate GitHub issues, write fixes, and open new pull requests based on a project-wide roadmap. Others monitor performance benchmarks and flag regressions on the group’s Discord server. Some agents, according to Decoder, even come to meetings and make requests to pull features from the conversation.
💜 for EU tech
The latest talk from the EU tech scene, a story from our genius founder Boris, and some incredible AI art. It’s free, every week, in your inbox. Register now!
The team also uses Clawpatch.ai, Vercel’s Deepsec, and Codex Security for bug analysis and additional security. The result is a development project where three people oversee an array of AI agents that collectively perform the work of what would otherwise be a centralized engineering team.
A question of cost
Steinberger has been outspoken about economics. He clarified that the sum of $1.3 million represents the Codex “Quick mode” price, which consumes credits at a much higher rate than normal usage. Disabling Quick Mode alone will reduce raw API costs to about $300,000 per month, a 70 percent decrease. At normal prices, the operation will still cost $3.6 million per year, but the gap between the number of headlines and sub-representers and how the base rate can put prices. increase the reported costs.
When asked about the return on investment, Steinberger said everything his team builds is open source and works with leading proprietary and open-source models. “I would say very high,” he said.
This figure is useful because vendor marketing around AI coding tools rarely reveals raw usage and token volumes at this scale. Most of the business teams that plan agent optimization tools work on assumptions and estimates. Steinberger’s bill is a concrete, public data center: 100 agents working continuously for 30 days on a large open source database costs between $300,000 and $1.3 million per month depending on execution speed, before any optimization.
Who is Peter Steinberger
Steinberger is not new to building engineering tools at scale. An Austrian developer founded PSPDFKit in 2011, starting a PDF editing and annotation framework that became the standard for mobile document management. By 2021, apps built on PSPDFKit were running on more than one billion devices worldwide, and the company raised $116 million from Insight Partners, its first outside investment after a decade of profitable, self-funding growth.
After leaving PSPDFKit, Steinberger began exploring AI agents as a personal project. OpenClaw, an autonomous AI assistant that runs entirely on user hardware, has become the fastest-growing open source project in GitHub history, surpassing 302,000 stars in April 2026, surpassing React, Vue.js, and TensorFlow in a fraction of the time those projects took to reach the same milestone. The framework connects to tools people already use, including email, calendars, browsers, and messaging platforms from Slack and Discord to WhatsApp and iMessage, and allows AI agents to issue shell commands, manage files, and automate web tasks locally.
When Steinberger joined OpenAI, he announced that OpenClaw would move to a private foundation to maintain its open source character. “I want to change the world, not build a big company,” he wrote. “Partnering with OpenAI is the fastest way to bring this to everyone.”
What is revealed about AI is the coding economy
The $1.3 billion bill comes at a time when the economy of AI-powered development is taking center stage in the software industry. OpenAI recently opened up ChatGPT subscriptions to 3.2 million OpenClaw users, allowing them to run private agents using the Codex endpoint for $23 per month. In contrast, Anthropic banned Claude Pro and Max subscribers from using OpenClaw and other third-party agent frameworks, concluding that the computing demands of independent agents making thousands of API calls per day were not economically sustainable under the low-rate subscription price.
The divergence between those two approaches reflects unresolved tensions in AI pricing. Subscription models are designed for human-speed interaction: a person typing questions in a chat window generates a manageable, manageable volume of API calls. Independent agent ships generate larger orders, and the gap between the subscription price and the actual inventory cost is a subsidy that the supplier or user pays.
Steinberger’s bill makes that gap visible. At $1.3 million for 100 agents, the cost per agent is approximately $13,000 per month, in addition to what the subscription system covers. Even with a $300,000 boost, each agent costs about $3,000 a month. For business teams exploring whether to use agency coding tools at scale, these numbers provide a foundation that no vendor marketing page will provide.
A broad pattern
OpenClaw’s trajectory, from a personal experiment to a highly-starred project on GitHub to a research platform sponsored by OpenAI, reflects a broad shift in the way software is built. AI coding agents from DeepMind, OpenAI, and Anthropic are moving from proof-of-concept demonstrations to production deployments, and the question is no longer whether AI will write significant amounts of code but how much it will cost and who will pay for it.
The rise of AI-assisted development, from coding pilots to private companies, is compressing the timeline between the aspirations of a three-person team and the outcome of a large engineering organization. Steinberger’s setup, three people and 100 agents, is an extreme version of what many companies will try on a smaller scale next year.
The $1.3 million bill is not a cautionary tale. It’s a receipt from the future, showing what it costs when AI development tools are used to their full potential, without the budget constraints that currently limit most teams to a fraction of what the technology can do. Whether that future is affordable depends on how quickly the cost of mathematical modeling falls, how well the orchestration frameworks for agents handle the use of tokens, and whether the security and quality challenges of AI-generated code can be handled at the speed these agents produce.
