janas
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 your RAM, pure C, CPU or GPU.
- avx2
- c
- cpu-inference
- gguf
- inference-engine
- linux
- llm
- local-llm
- mixture-of-experts
- moe
- nvme
- quantization
- qwen3
- simd
- speculative-decoding
- vulkan
- Stars
- 323
- Forks
- 20
- + today
- +5
- Created
- 24d
Ranking data as of October 13, 2026 (UTC).
Star History
Today, hour by hour
Overview
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 your RAM, pure C, CPU or GPU. It ranks #46 on GitTiger, gaining +5 stars on October 13, 2026 (UTC).
The project is written in C and has 20 forks. It was created 24d ago.
What it is
janas is a C project, licensed under GPL-3.0, described as an inference engine for running LLMs larger than system RAM on CPU or GPU. The README preview gives its status as experimental.
Capabilities described
- Performance claims, attributed to the description: 37 tok/s from Qwen3-Next-80B (48 GB) on a 32 GB laptop, 24 tok/s without drafts, against 2.8 for llama.cpp. - The description also claims it is faster than llama.cpp on all 13 models compared in Q4_K_M. These figures are the repository's own claims and were not independently verified. - The preview lists the test hardware and OS: Linux Debian 13, x86-64, and an Intel Core Ultra 9 185H with 32 GB RAM. - Topics include GGUF, quantization, mixture-of-experts, and NVMe.
Getting started
The supplied fields include no setup steps or commands. See the README at https://github.com/prabanta-dev/janas and the ChangeLog it references.
git clone https://github.com/prabanta-dev/janas.git
cd janas
# see README for setupFrequently Asked Questions
What is the reported speed?
The description reports 37 tok/s for Qwen3-Next-80B on a 32 GB laptop, against 2.8 for llama.cpp. These are the author's claims.
What hardware was used?
The preview lists Linux Debian 13 on an Intel Core Ultra 9 185H with 32 GB RAM. Other hardware was not supplied.
What license applies?
The metadata lists GPL-3.0.