Claude’s share of measured model R&D
Selected evidence from the State of AI Report 2026, with the period, source, and definitions needed to interpret it.
Evidence you can use
Claude’s share of measured model R&D
February-August 2026
| Period | Share led by Claude |
|---|---|
| February 2026Source | Under 1% |
| August 2026Source | 26% |
Anthropic’s internal index measures work inside one lab, under human supervision. It does not measure the share of all AI research automated or establish that AI can run a research agenda on its own.
Sources and dates
2026 report snapshot. Preview revised 2026-10-07. Individual data periods and source checks are listed below. This is not a claim that every source was updated on that date.
- Anthropic: measuring the pace of AI developmentFebruary-August 2026. Retained from the launch essay.
- Anthropic: recursive self-improvement2026. Retained from the launch essay.
- Karpathy: autoresearchRepository, changing over time. Retained from the launch essay.
- Nathan Benaich: Can AI learn scientific taste?2026. Author analysis.
- State of AI Report 2026, slide 6: Chinese open-weight models overtook American ones in AI research papers in 20262026 report snapshot. Read against the report PDF on 2026-10-07. Study-specific limits retained.
- State of AI Report 2026, slide 17: Can agents produce a top-tier research paper? No, but they can do its engineering.2026 report snapshot. Read against the report PDF on 2026-10-07. Study-specific limits retained.
- State of AI Report 2026, slide 23: As task benchmarks saturate, RSI evidence is moving inside the labs2026 report snapshot. Read against the report PDF on 2026-10-07. Study-specific limits retained.
- State of AI Report 2026, slide 36: But who benchmarks the benchmarks?2026 report snapshot. Read against the report PDF on 2026-10-07. Study-specific limits retained.
- State of AI Report 2026, slide 41: Long-horizon coding rankings change with the task and the evaluation budget2026 report snapshot. Read against the report PDF on 2026-10-07. Study-specific limits retained.
- State of AI Report 2026, slide 42: High scores can hide unfinished scientific analyses and desk work2026 report snapshot. Read against the report PDF on 2026-10-07. Study-specific limits retained.
- State of AI Report 2026, slide 43: METR needs harder tasks to reliably measure the strongest models2026 report snapshot. Read against the report PDF on 2026-10-07. Study-specific limits retained.
- State of AI Report 2026, slide 48: Generative video goes real time and lets a streamer steer it!2026 report snapshot. Read against the report PDF on 2026-10-07. Study-specific limits retained.
- State of AI Report 2026, slide 105: AI in education: the best tutor is not a helpful assistant2026 report snapshot. Read against the report PDF on 2026-10-07. Study-specific limits retained.
- State of AI Report 20262026 report snapshot. Report PDF. See individual slide references for the expanded analysis.
Cite this page
Benaich, Nathan. “Claude’s share of measured model R&D.” State of AI Report 2026. Published 2026-10-08.