METR: AI accelerates Anthropic model development by nearly twofold

An independent research organization, METR, reports that Anthropic’s model capabilities have grown 1.9 times faster since October 2025. Analysts attribute roughly 80% of this acceleration to the integration of internal AI tools into the company’s development workflows.

• Performance growth on key benchmarks is now 1.9 times faster.
• Approximately 80% of acceleration stems from AI-assisted research and development.
• Time between ideation and experiment has been reduced by roughly 80%.
• AI is beginning to handle complex tasks during the development lifecycle.

The impact of AI on R&D

The study accessed internal Anthropic data, including results from unreleased models (Models 2, 3, and 4), and involved interviews with 15 staff members. Claude has become a critical assistant for writing code, validating hypotheses, running experiments, and analyzing results. Engineers noted that the main bottleneck has shifted from manual coding to the availability of ideas and raw computing power.

While some employees subjectively estimated a 12-fold increase in personal productivity, METR warns that these are subjective reports rather than controlled experimental data. However, the qualitative evidence shows that Claude is effectively streamlining specific research stages, even if it has not yet fully automated the entire training process.

Transparency and ethical concerns

Anthropic allowed METR to study internal data but redacted significant portions to prevent competitors from gaining a roadmap for development. Co-founder Sam McCandless acknowledged the primary finding but admitted that the redacted data makes independent verification of the results difficult for outsiders.

This acceleration comes amid calls from CEO Dario Amodei to slow down AI development until safety measures catch up with model capabilities. METR's findings suggest that Anthropic may have already crossed its own early-warning threshold regarding the speed of development, sparking further debate on responsible AI scaling.

Strategic implications

The findings underscore a shift toward 'AI-native' development, where artificial intelligence is used to build its own successors. For global tech companies, this proves that integrating AI into R&D is no longer optional but a core requirement to stay competitive. The report predicts that no major technical hurdles will stop this acceleration within the next 12 months, though significant uncertainty remains.