The Million-Dollar Receipt: What OpenClaw Reveals About the Real Cost of AI Coding Agents

For most developers, the cost of AI is a monthly subscription or a few dollars in API credits. But for Peter Steinberger, the creator of OpenClaw, the price of pushing autonomous AI to its limit is far more stark. In a single 30-day period, Steinberger racked up a bill of $1,305,088.81 in OpenAI API tokens.

This expenditure isn’t just a curiosity of extreme usage; it’s a critical data point for the software industry. As companies transition from simple “copilots” to fully autonomous agent fleets, Steinberger’s bill serves as a “receipt from the future,” exposing the massive gap between flat-rate subscription models and the actual compute costs of agentic software development.

Breaking Down the $1.3 Million Bill

The scale of the operation was immense. To power his open-source project, Steinberger ran approximately 100 Codex instances simultaneously. Over 30 days, this setup generated 603 billion tokens across 7.6 million requests, primarily utilizing the GPT-5.5 model.

However, the headline figure includes a significant pricing nuance. Steinberger clarified that the total reflects Codex’s “Fast Mode” pricing, which consumes credits at a much higher rate than standard execution. By disabling Fast Mode, the raw API cost would drop to approximately $300,000 per month—a 70 percent reduction. Even at this optimized rate, the annual cost would sit at $3.6 million, highlighting that autonomous agents operating at scale are an expensive endeavor.

Replacing the Engineering Organization

The expenditure is justified by the sheer volume of work the agents perform. Rather than just writing snippets of code, Steinberger’s three-person team manages a pipeline where AI agents handle tasks typically reserved for a mid-sized engineering organization. These agents:

• Review pull requests and scan commits for security vulnerabilities.
• Deduplicate GitHub issues and write corresponding fixes.
• Open new pull requests based on the project’s roadmap.
• Monitor performance benchmarks and report regressions via Discord.
• Attend meetings and generate pull requests for features discussed in conversation.

To augment this fleet, the team integrates tools like Clawpatch.ai, Vercel’s Deepsec, and Codex Security for deeper bug and security analysis. The result is a hyper-efficient operation where a handful of humans oversee a massive automated workforce.

The Rise of OpenClaw and Peter Steinberger

Steinberger is no stranger to scaling developer tools. He founded PSPDFKit in 2011, a PDF rendering framework that reached over one billion devices by 2021 and secured $116 million in investment from Insight Partners after a decade of self-funded growth.

His latest venture, OpenClaw, has become the fastest-growing open-source project in GitHub history, surpassing 302,000 stars by April 2026. OpenClaw is a self-hosted autonomous AI assistant that allows agents to execute shell commands, manage files, and automate tasks across email, calendars, and messaging platforms like Slack, Discord, WhatsApp, and iMessage.

When Steinberger joined OpenAI in February 2026, he moved OpenClaw to an independent foundation to ensure it remained open-source, stating that teaming up with OpenAI was the fastest way to bring the technology to everyone.

The Great Subscription Divide: OpenAI vs. Anthropic

Steinberger’s bill exposes a growing tension in how AI companies price their services. Subscription models are designed for human-speed interaction, where a user’s queries are predictable. Autonomous agents, however, generate orders of magnitude more traffic.

This has led to two diverging strategies among the industry leaders:

• OpenAI: Has opened ChatGPT subscriptions to OpenClaw’s 3.2 million users, allowing them to run autonomous agents through the Codex endpoint for $23 per month.
• Anthropic: Has blocked Claude Pro and Max subscribers from using OpenClaw and other third-party agent frameworks, citing that the compute demands of thousands of daily API calls are economically unsustainable under flat-rate pricing.

The disparity is clear: at $1.3 million for 100 agents, the per-agent cost is roughly $13,000 per month. Even at the optimized $300,000 rate, each agent costs roughly $3,000 per month. This far exceeds any standard subscription fee, meaning the provider must either absorb a massive subsidy or move toward usage-based pricing.

Looking Ahead: The Future of Agentic Development

The $1.3 million bill isn’t a warning, but a benchmark. It demonstrates what is possible when budget constraints are removed—in this case, treated as a research investment by OpenAI. As AI coding agents from OpenAI, DeepMind, and Anthropic move into production, the primary challenge shifts from technical capability to economic viability.

Whether this model of development becomes accessible to the broader enterprise depends on three factors: the decline of model inference costs, the efficiency of agent orchestration frameworks in managing token usage, and the ability to maintain code quality and security at the breakneck speed these agents produce.