Anthropic analyzed 207,049 cases and published its findings on Claude usage.

Anthropic analyzed 207,049 Claude and Claude Code sessions tied to scientific research conducted between August 2 and September 30, comprising 128,914 Claude/Cowork sessions and 78,135 Claude Code sessions, the company said in a technical report titled "AI for scientific discovery: How researchers use Claude." Data analysis and writing accounted for the largest shares of session classifications, while the report's research-acceleration figures are conditional calculations rather than measured outcomes. The counts reflect sessions rather than users, and the Claude Code sample includes only interactive sessions, excluding programmatic usage. The findings matter as an early large-scale look at how researchers actually deploy AI assistants in day-to-day scientific work. What to watch next is whether Anthropic follows up with outcome-based metrics that can substantiate its acceleration claims.