Anthropic study of 207,000 research sessions
AI could take on 27–52% of researcher time per study (median 38%)
Confirmed
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 8 November 2026
Your AI and this story
- GPT-6 Astra189 days after its cutoff
- Claude Opus 5.5128 days after its cutoff
- Gemini 4 Argon219 days after its cutoff
- Grok 4.7158 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 128 days before it.
Key facts
- Next day (Nov 6) Anthropic’s Science Blog ran a guest essay by Jakob Macke, Daniel Gedon and Stefan Wahl (Tübingen) on how AI is changing scientific modelling, using Hodgkin-Huxley as a case study (X post ~427k views)
- Data: 128,914 Claude (Chat and Cowork) and 78,135 Claude Code research sessions, consumer plans only, classified by a model into the SciNet taxonomy (7,262 tasks, 34 fields)
- Lowest AI share in arts and anthropology, highest in statistics and computer science; lab sciences limited by physical work
- Assumptions include AI at $30/hour vs $95/hour for a researcher; scenarios ‘illustrative, not forecasts’
- Survey respondents recruited from Anthropic grant applicants (likely heavier AI users)
- Lead authors: Johannes Hermle, Ben Jones, Lukas Althoff
What happened
Quote: “AI could take on 27–52% of the researcher time needed to produce a new study, with a median of 38% across fields.” Physical experiments, equipment costs and idea quality are largely not modelled.
Sources
2 sources from 1 site. Numbers match the chips in the text.
2 sources: 2 primary
Primary
- Anthropic Science Blog: Teaching AI to build scientific modelsanthropic.com, official
- Anthropic: AI for scientific discovery — how researchers use Claudeanthropic.com, official
Changes
- Science Blog essay
- Filed