In a striking example of the costs associated with AI-assisted development, Austrian developer Peter Steinberger has reported spending over $1.3 million on OpenAI's API in a single month. This bill, totaling $1,305,088.81, covers 603 billion tokens processed across 7.6 million requests, driven by about 100 Codex instances managed by a small team working on the open-source OpenClaw project.
Steinberger, who started working with OpenAI in February, shared a screenshot of his API usage dashboard on social media, revealing the hefty expenses incurred. On the day of his post, his account showed a spending of $19,985.84 and 206,000 requests. The most utilized model during this period was GPT-5.5, released in April 2026. Despite facing challenges in the public eye, including controversies related to AI alignment and competition, Steinberger views OpenClaw as a laboratory for exploring AI's potential in software development without budget constraints.
Steinberger clarified that the $1.3 million expenditure was largely due to using Codex's "Fast Mode" pricing, which charges at a higher rate. He noted that disabling this mode could bring costs down to around $300,000, highlighting the significant difference in API pricing. For context, a single $200-per-month Codex Pro subscription typically offers around $5,000 to $6,000 worth of API credits, suggesting that Steinberger’s non-Fast Mode usage would equate to roughly 60 Codex Pro subscriptions.
Implications for AI Development Costs
The financial implications of such usage patterns raise important questions about the economics of AI development tools. OpenAI estimates that typical Codex costs range between $100 and $200 per developer per month, but Steinberger's usage exemplifies the extreme end of this spectrum. This contrast highlights the gap between what developers actually pay for these tools and the underlying computational costs incurred by providers.
As competition in the AI coding tool sector intensifies, with rivals like Claude Code and Cursor vying for developer adoption, the pricing models are under increasing scrutiny. All three platforms are known to subsidize inference costs to attract users, complicating the economic calculus for developers. OpenAI's recent shift to a token-based billing system has introduced greater transparency to these costs while also adding new variability for high-volume users.
Steinberger's reaction to the staggering bill is nonchalant; he sees it as a necessary expense for research into the significant potential of AI in software development. He emphasizes that all developments from the OpenClaw team remain open source, fostering the collaborative spirit of the project despite the financial strains.
A Look Ahead
As AI tools continue to evolve, the dynamics of pricing and usage will likely shift. The implications for developers and companies leveraging these technologies are significant. Steinberger's experience serves as both a cautionary tale and a potential roadmap for navigating the complexities of AI development expenses. As the industry progresses, balancing cost efficiency with the capabilities provided by AI will remain a critical consideration for all involved.
Quick answers
What is OpenClaw?
OpenClaw is an open-source project developed by Peter Steinberger that utilizes AI coding agents.
How much did Peter Steinberger spend on OpenAI’s API?
He spent $1,305,088.81 over the course of 30 days.
What is Codex’s Fast Mode?
Fast Mode is a pricing structure for Codex that consumes credits at a higher rate than standard execution.
Which model was primarily used during this period?
The primary model used was GPT-5.5.
