Anthropic's AI models are improving nearly twice as fast as before. A recent study by AI safety group METR has revealed this in a study on November 5. The group has found that the company’s model performance on certain tasks has been improving 1.9 times faster since October 2025. Researchers reached this finding after an 11-week study between August and October. They looked at private research, interviews and performance data from models that have not yet been released.
According to METR, “As of October 2026, Anthropic is likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025.”
Anthropic researchers have also reported major gains from using AI to write code and run tests. A few polls recently pointed out that productivity has increased around 12 times, although METR said these figures were not reliable.
If AI can help researchers build better models in less time, companies may not need the same level of human effort for every new step. However, AI is still unable to handle the most important stages of model training on its own.
AI is making the development to some extent faster and easier. Researchers can use AI to write code, test ideas, and complete tasks that used to take a long time previously. This could help companies save time and money while building better models. It may also increase competition, as companies look for ways to use AI to improve their own systems. However, fully independent AI development is still some distance away.
Also Read: Anthropic’s FlashProve to Verify Identity of AI Models
Faster progress also brings new risks. Safety testing, alignment research, independent reviews and government rules must keep pace with new AI systems. If companies build models faster than experts can test them, serious risks may go unnoticed. Researchers also warn that future AI systems could become harder to control and ensure they follow human goals. Regular safety checks and outside reviews will be important as these systems become more powerful.
Anthropic's findings show that AI is not just changing how people work. It is also changing how AI itself is built. The next challenge will be to make sure that safety and human control grow alongside this progress.