prabanta-dev/janas is a pure C project for running language models larger than RAM on a CPU or GPU. According to the GitHub release page, v2026.10.29.1 is a patch release following the previous day's release. This release includes Linux and Windows binaries and a graphics processing unit (GPU) diagnostic build. The changes are summarized in the repository's ChangeLog.md, but the provided information lists no specific fix items.
The repository description presents running models larger than memory and performance comparisons. The example given in the description is that running the 48 GB Qwen3-Next-80B on a 32 GB laptop recorded 37 tok/s. The case without draft is 24 tok/s, and llama.cpp is listed at 2.8 tok/s. Read speeds are 62 tok/s and 16 tok/s respectively. It also states that it is faster than llama.cpp for all 13 models compared in Q4_K_M format.
These figures and comparison results are claims recorded in the repository description. The provided material contains no measurement conditions or independent verification results. Therefore, the change that can be confirmed in this release is the availability of binaries for each platform and the GPU diagnostic build, and the performance advantage is best treated as a repository claim until separate supporting evidence is confirmed.