Skip to content
NetProjects
Menu

NetProjects

AI & Open Source

We port open AI models to Apple Silicon with MLX, check each port against the original, and publish the builds and code openly. Each port is an MLX build of a model someone else trained, not a model of our own. We compare its output with the original before we publish it, and we write up how it went in our Insights.

MLX builds, plus a few GGUF builds for llama.cpp, of open models from NVIDIA, IBM, H2O.ai, AutoTrust AI, Fastino, Tencent, SenseNova, Concept Country and Juneo Lab, kept current on Hugging Face and GitHub (see the collection). Not affiliated with any of them.

15 open AI models ported · 40 builds on Hugging Face

Counts as of 11 Nov 2026; downloads are all-time and refreshed daily (UTC). How we count

Every port, newest first

  1. Juneo Lab Laplace-1-Atlas 27B (MLX and GGUF)

    Juneo Lab's 27B decision model built on Qwen3.8-27B: give it a state, a question and a set of candidates, and it returns a probability for each candidate from one forward pass.

    Hugging Face
    6-bit8-bit4-bitbf16GGUF

    Not affiliated with Juneo Lab or Alibaba Cloud.

  2. Concept Country Qwen Concept 27B (MLX)

    Qwen3.8-27B with nine MLP down-projections edited by Concept Country so that it reaches its answer with less thinking; text and images.

    Hugging Face
    6-bit8-bit4-bitbf16

    Not affiliated with Concept Country or Alibaba Cloud.

  3. Tencent DSpine for Qwen3.8-27B (MLX)

    GitHub
    dspine-hybrid-mlx
    Hugging Face
    drafter
    Article
    Tencent's DSpine on a Mac: Qwen3.8-27B two to four times faster, same answers

    Not affiliated with Tencent, Alibaba/Qwen or Apple.

  4. NVIDIA Cosmos3-Edge reasoner and video generator (MLX)

    GitHub
    cosmos3-edge-mlx
    Hugging Face
    reasoner 8-bit6-bitVideo 8-bitVideo bf16
    Article
    NVIDIA Cosmos3-Edge on a Mac: a small model that understands video, and one that makes it

    Not affiliated with NVIDIA.

  5. SenseNova-Loop-9B-Preview (MLX)

    GitHub
    sensenova-loop-mlx
    Hugging Face
    8-bit6-bit4-bit
    Article
    SenseNova-Loop-9B on a Mac: a looped language model in MLX

    Not affiliated with SenseTime / SenseNova.

  6. NVIDIA Nemotron 3.5 Embed and Rerank (MLX)

    GitHub
    nemotron-retrieval-mlx
    Hugging Face
    Embed 1B 8-bitEmbed 8B 8-bitRerank 8B 8-bitRerank 8B GGUF
    Article
    NVIDIA's Nemotron 3.5 search models on a Mac: embeddings and reranking for text and page images

    Not affiliated with NVIDIA.

  7. Fastino GLiDE 27B (MLX)

    GitHub
    glide-mlx
    Hugging Face
    8-bit4-bit
    Article
    Fastino's GLiDE 27B on a Mac: a decision model that answers with probabilities, not text

    Built with Fastino GLiDE. Not affiliated with Fastino.

  8. IBM Granite Switch 4.1 8B and 30B (MLX)

    GitHub
    granite-switch-mlx
    Hugging Face
    8B 8-bit8B 4-bit30B 8-bit30B 4-bit
    Article
    Granite Switch 4.1 on Apple Silicon: chat, decisions and a safety judge from one model, in 8B and 30B

    Not affiliated with IBM.

  9. AutoTrust AI JEV-35B and AutoTrust-JEV-27B v1.1 (MLX)

    GitHub
    JEV-35BJEV-27B
    Hugging Face
    JEV-35B 4-bit mixedJEV-27B 8-bit
    Article
    Two AutoTrust AI decision models on a Mac: JEV-35B and JEV-27B on MLX

    Not affiliated with AutoTrust AI.

  10. NVIDIA Dory 1.0 4B (MLX)

    GitHub
    dory-mlx
    Hugging Face
    8-bitbf16NVFP4
    Article
    NVIDIA's looped Dory 4B model now runs on a Mac

    Not affiliated with NVIDIA.

  11. H2O-Lightning-4B (MLX)

    GitHub
    h2o-lightning-mlx
    Hugging Face
    8-bit4-bit
    Article
    H2O-Lightning-4B on a Mac: a calibrated decision model that knows when it isn't sure

Browse the builds collection on Hugging Face →

Related insights

netprojects on Hugging Faceinternet-projects on GitHubHow we use AI