DEVOURED
Independent Assessment of AI R&D Acceleration at Anthropic in October 2026
METR's independent assessment found that Anthropic’s AI R&D capabilities have accelerated by roughly 2x since late 2025.
What: Researchers at METR conducted a 11-week high-access evaluation at Anthropic, concluding that AI assistance in R&D is likely the primary driver of a doubling in capability improvement rates. The report highlights that compute scaling is not the cause of this acceleration.
Why it matters: This quantifies the 'AI-building-AI' feedback loop, validating concerns that AI adoption in research can lead to compounding acceleration in capability development.
Deep dive
- Anthropic models are growing on benchmarks at 1.9x the pre-October 2025 rate.
- The speedup is primarily attributed to AI-assisted coding and experimentation rather than fully autonomous R&D.
- No immediate bottlenecks in hardware or data were identified that would halt this trajectory in the next year.
- Anthropic exercised broad redaction rights over findings that could 'accelerate other AI developers'.
- The assessment relied on the Epoch Capabilities Index (AECI) to measure capability growth.
Decoder
- AECI (Epoch Capabilities Index): A composite benchmark metric used to track the progress of AI models on various tasks over time.
- Capability acceleration: A measurement of how much faster a model's performance improves compared to historical trends.
Original article
Full article content is not available for inline reading.
Explore all 54 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi