METR Report: Anthropic Model Benchmarking Progress Accelerates, Key Evidence Redacted. Independent model evaluation organization METR released an evaluation report stating that Anthropic's AI models progressed at about twice the speed in executing tasks with clear success criteria in October compared to the period between March 2024 and October 2025. The report suggests that AI assistance may be the primary driver behind this recent acceleration, but Anthropic redacted most of the core evidence supporting this conclusion. Anthropic co-founder Sam McCandlish responded that the redactions were made because the relevant details are commercially sensitive and could accelerate the progress of other developers.