TL; DR
OpenClaw creator Peter Steinberger spent $1.3 million in OpenAI API tokens in 30 days running 100 Codex instances on the open source project. The bill covered by OpenAI, which Steinberger is currently working on, represents 603 billion tokens across 7.6 million requests and provides the most concrete public data point on the value of autonomous AI coding at scale.
Peter Steinberger, founder of OpenClaw and an engineer at OpenAI, increased API spending by $1.3 million in one month by running about 100 Codex instances simultaneously on the open source project. The bill, which covered 603 billion tokens in 7.6 million requests over 30 days, is the clearest demonstration yet of what happens when AI-powered software runs without budget constraints, and how quickly costs can escalate when autonomous agents operate at scale.
Steinberger X posted a screenshot of the billshows $1,305,088.81 paid to OpenAI API with GPT-5.5 as base model. OpenAI covers costs: Steinberger joined the company in February 2026, and the costs are seen as a research investment to understand what software development looks like when the token economy is not a limiting factor.
Peter Steinberger
What agents actually do
100 Codex examples don’t just make code. Steinberger’s three-person team has built an autonomous development pipeline in which AI agents perform a number of tasks that would normally require a larger engineering organization. Agents review pull requests, scan a commit for security vulnerabilities, replicate GitHub issues, write fixes, and open new pull requests based on the project’s broader roadmap. Others track performance metrics and flag regressions on the team’s Discord server. Some agents, according to The Decoder, even attend meetings and create pull requests for features that come up in conversation.
💜 of EU technology
The latest rumblings from the EU tech scene, a story from our wise founder Boris and some questionable AI art. Free in your inbox every week. Register now!
The team also uses Clawpatch.ai, Vercel’s Deepsec and Codex Security for additional bug and security analysis. The result is a development operation where three people oversee a fleet of AI agents that collectively perform the work of a traditionally mid-sized engineering team.
A question of cost
Steinberger was transparent about the economy. He clarified that the $1.3 million figure was Codex’s “Fast mode” a price that consumes credit at a significantly higher rate than standard implementation. Disabling Fast Mode alone would drop raw API costs to about $300,000 per month, a 70 percent reduction. At standard pricing, it would still cost $3.6 million a year to operate, but how economically impressive the gap between the headline figure and the bottom line can be. Exaggerating reported costs.
When asked about the return on investment, Steinberger said everything his team builds is open source and works with leading proprietary models as well as open-weight alternatives. “I would say very high” he said.
This figure is very useful because vendor marketing around AI coding tools rarely discloses raw costs and token volumes on this scale. Most enterprise teams planning agent development tools work based on projections and estimates. Steinberger’s bill is a concrete, public data point: 100 agents running 30 days non-stop on a large open-source codebase costs between $300,000 and $1.3 million per month, depending on execution speed, before any optimization.
Who is Peter Steinberger?
Steinberger is no newcomer to building developer tools at scale. An Austrian engineer founded PSPDFKit in 2011, uploading a PDF rendering and annotation framework that has become the standard for mobile document processing. By 2021, apps built on PSPDFKit were running on more than a billion devices worldwide, and the company raised $116 million from Insight Partners, its first outside investment after a decade of profitable, self-funded growth.
After leaving PSPDFKit, Steinberger began experimenting with artificial intelligence agents as a personal project. It became OpenClaw, a self-driving autonomous AI assistant that runs entirely on users’ own hardware. The fastest growing open source project in GitHub historyBy April 2026, it surpassed 302,000 stars, surpassing React, Vue.js, and TensorFlow in a fraction of the time it took those projects to reach similar milestones. 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 execute shell commands, manage files, and automate web tasks natively.
When Steinberger joined OpenAI, he announced that OpenClaw would transition to an independent foundation to maintain its open source nature. “I want to change the world, not build a big company.” he wrote. “Teaming up with OpenAI is the fastest way to make this available to everyone.”
What AI reveals about the coding economy
The $1.3 million bill comes at a time when the economics of AI-powered development are a central preoccupation of the software industry. OpenAI recently opened up ChatGPT subscriptions to OpenClaw’s 3.2 million usersallowing them to manage autonomous agents through a Codex endpoint for $23 per month. Anthropic, on the other hand, blocked Claude Pro and Max subscribers from using OpenClaw and other third-party agent frameworks, concluding that the computing demands of autonomous agents handling thousands of API calls per day were not economically sustainable under flat-rate subscription pricing.
The difference between these two approaches reflects an unresolved tension in AI pricing. Subscription models are designed for human-speed interaction: a person typing a request into a chat window generates a predictable, manageable volume of API calls. An autonomous agent fleet generates larger orders, and the difference between the subscription price and the actual computing costs is either a subsidy received by the provider or paid by the user.
Steinberger’s bill makes that loophole visible. At $1.3 million for 100 agents, the cost per agent is approximately $13,000 per month, which covers any subscription plan. Even with $300,000 optimized, each agent costs about $3,000 per month. For enterprise teams evaluating whether to deploy agent coding tools at scale, these numbers provide a baseline that no vendor’s marketing page can offer.
A wider model
OpenClaw’s trajectory reflects a broader shift in how software is built, from a personal experience to a top-star project on GitHub and an OpenAI-sponsored research platform. AI coding agents from DeepMind, OpenAI and Anthropic The concept is moving from proof-of-concept demonstrations to production deployment, 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-powered developmentfrom individual coding pilots to fleets of fully autonomous agents, it compresses the timeline between the ambitions of a three-person team and the product of a large engineering organization. Steinberger’s setup of three people and 100 agents is an extreme version of what many companies will try on a smaller scale next year.
A $1.3 million bill is not a cautionary tale. It’s a future receipt that shows what AI development tools cost when used to their full potential, without the budget constraints that currently limit most teams to a fraction of what the technology can do. Whether the future is viable depends on how quickly model extraction costs decrease, how effectively agent orchestration frameworks manage token usage, and Security and quality issues of AI-generated code can be controlled by the rate at which these agents are produced.
