Janas: Run LLMs larger than RAM by streaming experts from SSD

What changed

Janas-LLM allows running language models larger than system RAM by streaming experts directly from an NVMe SSD, outperforming llama.cpp on tested hardware.

Why you should care

Streaming weights from SSD enables running large language models on resource constrained consumer hardware.

Your move

Test. Verify performance claims on local development machines.

Evidence

Unverified sourceHacker News — LLM ↗

What to watch next

Independent benchmarks confirming long term stability and hardware wear metrics for SSD streaming.

Event

release

Event date

Oct 12, 2026

Relevant to

General AI readers