Anthropic Claude 26 Percent RD

Anthropic published a number this week that should reset how fast you think AI capability is moving: Claude now drives 26% of the company's own research and development.

At the start of this year, that figure was effectively zero. Roughly 90% of Anthropic's R&D now happens in active collaboration with Claude, with the model completing large chunks of work end-to-end from a high-level prompt, still under human supervision, but doing far more of the heavy lifting than most people assume.

Anthropic is framing this explicitly as a way to track progress toward recursive self-improvement: an AI system capable of autonomously building its own successor. They're not claiming to be there. They're claiming to be measuring the distance, publicly, on purpose.

That transparency is itself notable. Most labs treat this kind of internal velocity metric as a competitive secret. Publishing it is a bet that showing the curve builds more trust than it costs in hype.

For anyone planning AI strategy on a 12-to-24-month horizon, this is the number to watch, not the next model release. A lab whose own tools are compounding its R&D speed is not on the same timeline as one that isn't.

What does a 26%-and-climbing number like this change about how you're planning for next year?