Anthropic says its internal Claude use, at list price, exceeds staff pay: Reveneau

Why it mattersThe vendor is treating its own token use as a line that can be compared to payroll. A team using Claude or Codex can measure the same ratio for its own month.

A model company has told the SEC that its own tool now costs it more than its people. The claim is in Anthropic's amended Form S-1. Priced at list, the value of its internal Claude use passed its monthly staff pay in April 2026. Stock pay is excluded from that comparison.

Business Insider surfaced the line this week in a piece on Anthropic's IPO filing. The filing itself gives the growth number. Internal Claude use at list prices has grown over 250 times from January to September 2026. Both figures come from Anthropic's own words.

What the filing says about Claude writing Anthropic's code

The S-1 pairs the pay comparison with a second number from the same section. Over 90 percent of the contributions to Anthropic's codebase in September were written by Claude, in the company's own account.

The filing also gives a per-engineer figure. Daily code per engineer in the three months to September was over 45 times the daily average for 2024. In the same twelve months, trailing revenue grew about 1,200 percent. Headcount grew about 170 percent. Anthropic ties these numbers together. Claude writes most of the code, each engineer produces much more code than before, and revenue grows ahead of hiring.

What the retail-price framing hides

Anthropic is not spending more cash on Claude than on salaries. It is counting its own token use at the price it charges outside customers. The S-1 is explicit: "at list prices." Business Insider makes the same point in its headline.

That accounting choice matters. A list price is a sticker price. Anthropic sets it. Running its own model for its own staff costs it only the compute. The cash figure, if the filing had given one, would be smaller. The company picked the retail number because it compares cleanly to payroll, which is also at list.

A reader looking at the same ratio for their own team has three options. Price internal tokens at list, at cost, or at a transfer price between teams. The three answers can differ by an order of magnitude, and only the first is what Anthropic is reporting here.

What is left after the framing argument

Even with the framing stripped out, the twelve-month numbers in the filing stand on their own. Revenue grew about seven times faster than headcount. Daily code per engineer went up 45 times. Over 90 percent of the September codebase additions came from the company's own model.

Those are the figures a team can test against its own month. Count the tokens, apply the vendor's list price, divide by the payroll for the engineers who used the tool. The resulting ratio has every reason to be smaller than Anthropic's. The direction of the shift, from paying people to paying for tokens, is the same at any size.

Source

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.