Overview
- METR published an 11‑week, third‑party assessment on Thursday that estimated Anthropic’s benchmark progress has accelerated about 1.9 times since October 2025.
- The evaluators concluded roughly 80 percent of the measured speedup is likely driven by AI assisting research and development rather than by faster growth in training compute.
- An unreleased internal system called Model 4 scored above 50 percent on every METR benchmark, the first model in METR’s dataset to clear that threshold.
- METR’s findings are limited because Anthropic redacted much of the most informative evidence and the acceleration is measured mainly on well‑specified benchmarks that may not reflect messy real‑world tasks.
- Anthropic said it will publish its own analysis of development pace and its Responsible Scaling Policy in the coming weeks, a disclosure that could shape investor, customer, and regulator reactions ahead of a possible November IPO.