OpenClaw and the $1.3 Million OpenAI Bill

Imagine waking up to a bill for over a million dollars. Most of us would faint on the spot. But for Peter Steinberger, it's just another Tuesday at the office.

He leads a tiny team of three people. They run the OpenClaw project. Last month, their bill for OpenAI tokens hit $1.3 million. That is a massive amount of cash.

It's not just a number on a screen. It is a new way to write code. We are watching a high-stakes test of how far AI can go.

When the code writes itself

Steinberger joined OpenAI earlier this year. He brought his OpenClaw project with him. Now, he has access to a huge fleet of AI agents. These agents do not just chat. They work.

They handle the boring stuff. They review pull requests and scan for security holes. They even fix bugs on their own. It is an army of digital workers. They don't sleep or take breaks.

The team uses about 100 Codex instances. These agents monitor everything. They even join meetings to talk about new features. Then, they write the code for those features immediately.

The project acts like a lab. Steinberger wants to know what happens without limits. He ignores the budget to push the tech. It is a bold move for any developer.

The math behind the madness

The numbers are wild. We are talking about 603 billion tokens. That happened over 7.6 million requests. The model used was GPT-5.5. It is clearly a beast.

Steinberger posted a screenshot of his dashboard. It showed a spend of $1,305,088.81. This covers just 30 days of work. A single day cost almost $20,000.

He says "Fast Mode" is to blame. This mode runs tokens at a premium rate. If he turned it off, the bill would drop. He estimates it would land around $300,000.

Even that lower price is huge. It costs more than 60 Codex Pro subscriptions. This shows the gap between consumer pricing and real compute costs. We are seeing what this tech really eats.

What the bits and bytes mean

The tech world is watching closely. Costs are the biggest hurdle for AI tools. Codex, Claude Code, and Cursor all fight for users. They subsidize the costs to keep us hooked.

When OpenAI moved to token billing, it changed things. It made the costs clear to everyone. Now, power users see the true weight of their usage. It is not cheap to automate everything.

Some people think this is a waste. They argue that humans could do the work for less. Hiring 70 senior engineers would cost about the same. Can 100 agents beat 70 pros? That is the real question.

Steinberger doesn't seem worried. He thinks this is the future of building software. If you remove the cost, how fast can you move? He is trying to find out.

The long road ahead for AI devs

We are just at the start of this shift. Companies are betting big on agentic workflows. They want to cut time to market. They want to ship code faster than ever.

Will these costs come down? Probably. New chips and better models will help. But for now, it is an expensive experiment. It is a playground for the few.

Most teams can't afford this. They have to play it safe. They use agents for small tasks. They don't let them run the whole shop.

This project is a look into the future. It's messy, loud, and incredibly pricey. But it is also a sign of what is coming next.

Quick questions answered

Is this bill real? Yes. Steinberger shared the dashboard screenshot. OpenAI covers the cost since he is an employee.

What is OpenClaw? It is an open-source project. It uses AI agents to automate software development tasks.

Why is the bill so high? High-intensity agent usage burns through tokens. Using "Fast Mode" also increases the cost per token.

Can normal devs do this? Not easily. The cost is too high for most. You would need a massive budget to replicate this setup.

Is this sustainable? Maybe not yet. The goal is to learn from the usage. They are stress-testing the limits of the tech.