# Janas reports Windows tests of its disk-streaming language-model engine

_Published Tuesday, October 13, 2026 at 3:19 AM EDT · AI · Latest · Tier 2 — Notable_

![37 tok/s from Qwen3-Next-80B (48 GB) on a 32 GB laptop, 24 without drafts vs 2.8 for llama.cpp; reading 62 vs 16 tok/s. Faster than llama.cpp on all 13 models compared in Q4_K_M. LLMs bigger than y... — Primary](https://opengraph.githubassets.com/f2f187b2269a5977591f7a742789796c7319cfd0013828827ec593c65e77a6de/prabanta-dev/janas)

The experimental Janas-LLM project reports Windows 11 tests on October 12 showing that its language-model engine matched Linux generation speeds on the same laptop. It generated 25.6 tokens per second with Qwen3-4B and 34.2 with Qwen3-30B-A3B.

Janas runs large models by keeping weights on disk, streaming the expert components needed for each token from an NVMe SSD and caching frequently used components in memory. On Windows, Qwen3-4B processed prompts at 346 tokens per second against 193 on Linux. The measurements come from one Intel Core Ultra 9 185H laptop with 32 GB of RAM; other hardware configurations remain untested.

## Sources

- [github.com](https://github.com/prabanta-dev/janas)

---
Canonical: https://techandbusiness.org/newswire/NHE_wq-LpmUFYUvQHtGsPc
Published: 2026-10-13T07:19:18.947Z
Story chronology: 2026-10-12T00:00:00.000Z
Retrieved: 2026-10-13T09:40:18.034Z
Publisher: Tech & Business (techandbusiness.org)