Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology

Moonlight-16B-A3B

Model basics

Reasoning traces

Not supported

Context length

8K tokens

Max output length

No data

Model type

聊天大模型

Release date

2025-02-23

Model file size

32GB

MoE architecture

Total params / Active params

160.0B / N/A

Knowledge cutoff

No data

Inference modes

No mode data

Moonlight-16B-A3B

Open source & experience

Code license

MIT License

Weights license

MIT License- 免费商用授权

GitHub repo

https://github.com/MoonshotAI/Moonlight

Hugging Face

https://huggingface.co/moonshotai/Moonlight-16B-A3B

Live demo

No live demo

Moonlight-16B-A3B

Official resources

Paper

Muon is Scalable for LLM Training

DataLearnerAI blog

月之暗面开源了一个全新的160亿参数规模的MoE大语言模型Moonlight-16B：其训练算力仅需业界主流的一半

Moonlight-16B-A3B

API details

API speed

No data

No public API pricing yet.

Moonlight-16B-A3B

Benchmark Results

No benchmark data to show.

Moonlight-16B-A3B

Publisher

Moonshot AI

View publisher details

Moonlight-16B-A3B

Model Overview

月之暗面开源的一个160以参数的混合专家大模型，每次推理参数共30亿。效果超过同类型的大模型。

本版本是其基座的版本。

关于Moonlight-16B-A3B模型的详细介绍参考DataLearnerAI的官方博客： https://www.datalearner.com/blog/1051740316091143

DataLearner 官方微信

欢迎关注 DataLearner 官方微信，获得最新 AI 技术推送