Can I run Qwen 3 8B?

Qwen 3 8B is the model you reach for when you want a local assistant that can actually think a bit, not just chat. At 4-bit it comes in around 5 GB, so it fits on an 8 GB GPU, and the 8 GB minimum RAM figure means a base Apple Silicon Mac or a modest desktop can host it without drama. The headline feature is its hybrid thinking mode: it can spend extra tokens reasoning through a problem before answering, which is why people pick it over plain chat models in the same 8B class.

Day to day it feels quick. On an RTX 3060 12GB you get roughly 62 tokens per second, on an M-series Max around 70, and an RTX 4090 pushes it to about 172, all fast enough that responses outrun your reading. CPU-only on DDR5 drops to around 10 tok/s, fine for batch work but not interactive. The 128K context is genuine, but be careful: filling it all the way pushes total memory to about 22 GB, far past what an 8 GB card holds, so keep working context modest unless you have headroom or drop to the 3.4 GB q2 build.

Against Granite 3.3 8B, which is a similar-size open model aimed mainly at chat, Qwen 3 8B generally has the edge on multi-step reasoning and math thanks to its thinking mode, while Granite tends to feel more conversational. Its standout trait is that reasoning toggle in a package this small. And the license is the easy part: Apache 2.0 means you can use it commercially and in production with no provider-specific strings attached, which is rarer than it should be at this size.