Report: Anthropic Trains Claude AI at $4 an Hour, Outperforming $150/Hour Human Researchers

Anthropic is training Claude AI at a cost of roughly $4 per hour, with performance that outperforms human researchers billing at $150 an hour, according to a report from Chinese tech outlet 36 Kr.

The report, circulated via Google News, frames the efficiency gap as a striking illustration of AI's economics on research-style work: at $4 versus $150 an hour, the marginal cost of the AI is less than one-thirtieth of the human rate.

The claim is notable because Anthropic is both the model's developer and its first heavy user. Using its own Claude for tasks previously handled by human researchers gives the company a direct way to compress internal costs while demonstrating the model's capability.

If the figures hold, they add a concrete data point to the ongoing debate over AI displacing knowledge workers, with implications for staffing structures, the research outsourcing market, and how frontier labs price their models.

For Anthropic, the internal deployment also doubles as a credibility signal: a frontier lab trusting its own model for demanding research work is a stronger endorsement than any benchmark chart.

Watch for whether Anthropic discloses details of the program, how far Claude's role extends beyond human researchers, and whether rivals adopt similar cost-cutting deployments.