Monday, October 12
7 new models on Hugging Face that day.
-
Chat and writingLiquidAI
LFM2.5-24B-A2B
A Liquid AI mixture-of-experts language model with 24B parameters, of which about 2.45B are used per token.
- Covers 16 languages, including Arabic, English and Chinese
- 131K-token context; the card reports 184 tokens per second on an Apple M5 Max
- The card recommends temperature 0.1, top-k 50 and repetition penalty 1.05
Minimum 16GB graphics card e.g. GeForce RTX 5060 Ti 16GBor a computer with 32GB memory and no graphics card (slower)
Check the licence before useSupports Arabic
Download -
Reads imagesnvidia
NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16
An NVIDIA model that reads images and video along with text, with 120B parameters of which about 12B are used per token.
- Takes images and MP4 video up to 2 minutes, with a context of up to 1M tokens
- A hybrid of Mamba-2 and Transformer layers with a mixture of experts
- Video support depends on how you run it; some setups take only text and images
Minimum Workstation with 96GB of graphics or unified memory
Free for commercial use
Download -
Writes down speechnineninesix
tyger
652M parameters
Minimum A recent phone or any computer
Free for commercial use
Download -
Makes videoSkywork
Matrix-Game-4.0
Minimum 32GB graphics card e.g. GeForce RTX 5090
Free for commercial use
Download -
Makes videoVAST-AI
EdenLive
Minimum Server with several graphics cards
Check the licence before use
Download -
Speaks textASLP-lab
SemBridge
818M parameters
Minimum A recent phone or any computer
Free for commercial use
Download
The minimum is our estimate from the model’s size: a 4-bit copy for text models, 8-bit for image, video and sound. Licences and sizes come from each maker’s page; read the licence before using a model in your work.
