Small Qwen models released!!
it's not a matter of if, it's a matter of qwen.
Qwen (Alibaba’s AI model series) released their ‘Qwen 3.5 Small’ series today. They come in a few sizes:
Qwen3.5-0.8B and Qwen3.5-2B: These tiny models are designed for edge inference, like phones or smaller devices.
Qwen3.5-4B: This model can be used on entry level laptops.
Qwen3.5-9B: This one focuses on reasoning and agentic workflows. I was especially impressed with this one!
Basically the B stands for ‘billions of parameters’, which more or less indicates the size of an LLMs brain (similar to the number of neurons).
Models like GPT 5 or Claude Opus 4.6 are rumoured to be ~2 trillion parameters in size and require massive datacenters for hosting, whereas these Qwen models are designed to be run on consumer devices.
Below are some of their benchmarks:
It’s pretty mindboggling to see a 9B parameter model outperform GPT-OSS 120b, OpenAI’s open source model released in only August 2025. This comes via heaps of improvements to model architecture, better data, varied reinforcement learning pipelines etc.
Open source models have quickly caught up to the leading labs, and beg the question of whether a ‘moat’ exists for companies at the model layer. I’ll dive into this further in another post though. I’d definitely recommend using LMStudio to try out these models locally, the 2B one should run on basically any device!
Let me know if you guys have any questions about running models locally or in general, and hope you enjoyed. Don’t forget to like and subscribe!!