Скачать или смотреть Qwen 3.5 Small Series Models Overview - Tested on The M3 MacBook Air & 16GB Raspberry PI w/ Openclaw

Alibaba just dropped the Qwen 3.5 Small Series, four open source models (0.8B, 2B, 4B, and 9B parameters) that are punching way above their weight class. In this video, I break down what makes this release different: a hybrid architecture using Gated Delta Networks + Mixture of Experts, native multimodality (not bolted on), 262K context window, and benchmark numbers that have no business coming from models this small.

The 9B model is outperforming last-gen 30B+ models and going toe-to-toe with GPT-oss-120B on reasoning, math, and vision tasks. The 4B handles text, images, and video from ~8GB of VRAM. And the 0.8B? It runs on a phone.

All models are Apache 2.0 licensed and available on Hugging Face and ModelScope right now.
🔗 Links & Resources:
Qwen 3.5 on Hugging Face: https://huggingface.co/Qwen
Qwen 3.5 GitHub: https://github.com/QwenLM/Qwen3.5


This is part of my ongoing coverage of the open source AI model space. If you're a developer trying to figure out which models are actually worth running locally, subscribe, I'm testing and comparing these head-to-head.

#Qwen #Qwen3.5 #OpenSourceAI #LocalAI #SmallModels #AI #LLM #OnDeviceAI #MacMini #RaspberryPi #openclaw