Anthropic: computer science dominates research use of Claude

Computer scientists account for close to a third of Claude Code sessions tied to research even though they make up only 4.3% of researchers worldwide, according to a new Anthropic analysis of how scientists use the company's models.

The paper, "AI for Scientific Discovery: How Researchers Use Claude", looks at 207,049 consumer-plan sessions across Claude, Claude Code and Claude Science between August 2 and September 30, 2026, classifying each one against the SciNet taxonomy of scientific fields and tasks. "Computer science alone accounts for 32.7% of Claude Code sessions," the report finds, with AI and machine learning showing session shares of 5.0% for Claude and 15.2% for Claude Code.

What researchers actually do with the models skews toward the keyboard-bound parts of science. "Data analysis accounts for 33.7% of sessions on Claude and 48.1% on Claude Code," the report says, and writing and communication tasks "show high usage, especially on Claude (32.0%, compared with 16.0% of researcher time)." Ideation and hypothesis generation, which researchers themselves rank as more important, come in at 4.2% of Claude sessions and 2.1% of Claude Code sessions.

The report also runs scenarios on how much the tools could speed research up. At current capabilities, Anthropic estimates accelerations "ranging from 1.35x in the arts...to 1.96x in computer science and 2.02x in statistics." Under a full-automation scenario the gap widens: "statistics, computer science...could reach about 15x," while "bench sciences...stay near 3–4x" because the physical work of running experiments still has to happen in a lab.

On what researchers expect of the technology itself, the report cites survey responses showing "about 40% of respondents expect AI to carry out full research projects by 2030" and that "94% expect AI will match the best researchers in data analysis" by the same year.