OpenClaw creator – OpenClaw’s creator, Peter Steinberger, posted spending screenshots showing his CodexBar token usage nearly $20,000 in a single day and $1.3 million over 30 days on OpenAI’s API. While he says the “perks” cover the cost because he now works at OpenAI, social me
A screenshot posted to social media set off a wave of alarm among AI watchers: OpenClaw creator Peter Steinberger shared his CodexBar token spending, including nearly $20,000 in tokens used on OpenAI’s API in a single day.
The bigger figure landed days later.. In 30 days, Steinberger said he spent $1.3 million worth of tokens on OpenAI’s API, based on the image.. In the same thread. he emphasized he doesn’t have to pay out of pocket because he now works at OpenAI. writing that the token funds were “perks of OpenAI supporting OpenClaw.” When a user asked if OpenAI charged him for tokens. he replied. “ofc not.”
For many commenters, the number immediately reframed what “intensive use” of AI can cost.. The spending was treated as emblematic of a broader shift as access to free compute becomes a selling point in the competition for AI talent.. Others pointed to how “tokenmaxxing culture” has spread in Silicon Valley and referenced competitive “token leaderboard” behavior tied to multiple companies.
Still, the reaction was sharply personal once the monthly total was in view.. Several users said they were stunned to see over $1 million in AI spending for one engineer. with one comment claiming. “Someone’s burning through enough tokens to bankroll a small startup.” Others questioned whether that money could instead cover salaries for new engineers.
The tension inside Steinberger’s own response was that he insisted the work is “lean.” He wrote that the majority of his spending goes toward developing OpenClaw. a viral AI agent that he says sparked a Mac Mini buying craze and has since become the fastest-growing open-source product of all time.. He also shared a screenshot of his GitHub showing dozens of projects. ranging from a device sleep tool to a system for AI agents to make phone calls.
Some users pushed back directly on the “lean” claim after seeing the $1.3 million number.. Steinberger later acknowledged on social media that people were “freaking out over my AI spend. ” saying he was trying to answer a question: “How would we build software in the future if tokens don’t matter?” He then listed where tokens were going: AI agents that listen to meetings and begin working based on what he says. plus agents that review comments for spam.. “All that automation allows us to run this project extremely lean. ” he wrote—an assertion that drew dispute in the comments. where some said a project spending $1.3 million a month couldn’t be “lean.”
There was also doubt about whether the number shown was actually his usage. On his original post, a user asked whether the number pictured was his usage. Steinberger answered, “Yup! At least on this account.”
His exchanges also included an argument about whether the screenshots might function as marketing. “You shipped nothing,” one user wrote, and Steinberger shot back, “You seem to have a very particular definition of nothing.”
The relationship between these posts is hard to miss: Steinberger pairs a massive token total—$1.3 million in 30 days—with the insistence that the spending is covered through “perks of OpenAI supporting OpenClaw. ” and he links that automation to a claim of running the project “extremely lean.” At the same time. the same thread shows commenters questioning the “lean” framing. asking whether the figure could be bigger. and challenging whether the screenshots serve a purpose beyond engineering.
With the screenshots and replies now circulating. the debate has narrowed to one practical question: if token use can reach seven figures in a month for a single builder. and those costs can be offset by workplace perks. what does “efficient” software development look like when compute access stops being a personal expense.
OpenClaw Peter Steinberger CodexBar AI tokens OpenAI API token spending AI agents tokenmaxxing open-source Mac Mini