What Do You Need to Run Qwen Locally? VRAM by Version

For most versions, yes. Qwen3 8B, Qwen3 32B, Qwen3.5 9B, Qwen3.6 27B, Qwen3.6 35B A3B and Qwen3.8 27B are all Apache 2.0, so you can use them commercially, fine-tune them and redistribute them, as long as you include the license, keep the copyright notices and mark any files you changed. There are two exceptions, and both attach conditions to commercial use. Qwen3.8 Flash Next uses the Qwen Community License 1.0. If you use it in a commercial product with more than 100 million monthly active users or monthly revenue above the threshold, you must prominently display the model name in the interface. In addition, any company (including its affiliates) that runs a model-as-a-service business, or builds a standalone assistant product focused mainly on AI coding or office work, must apply to Qwen for a separate license before commercial use, regardless of size. Qwen3.8 2.4T A95B uses the Qwen3.8-Max License, which has the same attribution requirement, but its separate-license clause applies only to companies in those two lines of business whose revenue over 12 consecutive months exceeds the threshold. Both licenses state that the separate-license clause doesn't apply if you use the model only internally and don't offer its capabilities to third parties.

Every version except the 2.4T flagship has official tags: qwen3:8b, qwen3:32b, qwen3.5:9b, qwen3.6:27b, qwen3.6:35b-a3b and qwen3.8:27b. The Qwen3.8 Flash Next library has no latest tag, so you have to write the tag out in full, such as qwen3.8-flash-next:125b-a6b-q4_K_M. Ollama names it 125B because it doesn't count the separate 51B of lookup tables; the 176B figure on this site includes them. Qwen3.8 2.4T A95B has neither a local nor a cloud version on Ollama. To run it, use llama.cpp with unsloth's GGUFs, or run the official weights on vLLM.