Linus torvalds introduces first ai-assisted kernel patch with 3x performance boost

Linux creator Linus Torvalds has released his first kernel patch developed with the aid of a large language model (LLM), achieving a threefold increase in file read throughput. The patch, named 'noref-filemap-read-with-llm', was pushed to his personal Git repository on October 13, 2026. It addresses a long-standing bottleneck in Linux memory management by improving scalability for page reference counting. Torvalds confirmed that the core code, test programs, vulnerability reviews, and initial documentation were all developed with the help of Anthropic's Claude Code. In testing, the patch increased read throughput from 9.9 to 30 million operations per second, while reducing kernel clock cycles per read from 2,950 to 725. The patch remains in an experimental branch and has not been merged into the Linux 7.4 mainline. During recent events like the Maintainers Summit and code cleanup in the assoc_array repository, Torvalds also used LLMs to refactor outdated code and uncovered three long-standing bugs, some of which had been present since 2013.