OpenClaw Creator Accumulates $1.3M in AI Tokens

Peter Steinberger, the developer behind the open-source project OpenClaw, posted an API-usage screenshot that shows $1,305,088.81 charged to the OpenAI API over a 30-day window, according to reporting by The Next Web and Tom's Hardware. The bill reportedly covered 603 billion tokens across 7.6 million requests while running roughly 100 instances of Codex and logging about 206,000 requests on a single day, per Tom's Hardware and The Decoder. The dashboard listed GPT-5.5 as the top model, per Tom's Hardware. Multiple outlets report that OpenAI is covering the cost and that Steinberger, who joined OpenAI in February 2026, described the token access as a perk of OpenAI supporting OpenClaw, according to Business Insider and The Next Web. Business Insider and The Next Web also cite Steinberger's X posts, including the reply "ofc not" when asked whether he was paying the bill.

Per Tom's Hardware and The Decoder, Steinberger clarified in a follow-up post that the public $1.3 million figure reflects Codex "Fast Mode" pricing and that turning off Fast Mode would drop raw API cost to roughly $300,000. Industry-pattern observations: persistent agent fleets amplify token consumption in two ways - continuous background polling/monitoring and repeated short-turn interactions - which can produce large aggregate token counts even when per-call costs are modest. For autonomous code-assistant workloads, agents that attend meetings, review pull requests, run security scans, and open PRs multiply request rates compared with human-only processes; practitioners building similar systems should treat sustained high-frequency calls as a predictable cost driver rather than a one-off spike.

Reporting by Business Insider and Newser frames the disclosure inside a broader "tokenmaxxing" culture and competitive token leaderboards in the Valley, where free or subsidized API access is used as a recruiting or retention perk. Observed patterns in similar transitions: when companies or projects experiment with unconstrained compute budgets, they often uncover new workflows and automation benefits but also surface previously hidden operating expenses. Publicly visible, subsidized tabs create an outsized data point for the cost of agent-led development, but they do not by themselves reveal long-term unit economics, amortization, or the value captured by downstream productivity gains.

• •Whether other teams 공개ly report comparable token metrics for multi-agent deployments, which would help normalize per-agent cost estimates.
• •Vendor pricing signals: changes to model pricing, Fast Mode availability, or new flat-rate plans from APIs that could alter economic trade-offs for persistent agents.
• •Open-source alternatives and on-prem options gaining traction as teams factor token burn into procurement.
• •Any clarifications from OpenAI about how the company classifies or budgets subsidized research spending, which outlets like The Next Web describe as treating the tab as a research investment.

This episode provides a concrete public data point - 603 billion tokens and 7.6 million requests in 30 days - for estimating upper-bound costs of always-on agent pipelines. Industry context: teams considering fleets of specialized agents should instrument request patterns, evaluate lower-cost execution modes (for example, disabling Fast Mode), and model agent concurrency to forecast monthly token budgets more realistically.