Model family · 8 sizes
Qwen3: which size runs locally?
Qwen3 comes in 8 sizes, from 0.6B to 235B. Its strongest tracked here, Qwen3 30B-A3B, scores an LMArena Elo of 1383. Here is each size with its Q4_K_M weight, the memory it needs, and the hardware that runs it.
- Sizes
- 8
- Smallest
- 0.6B
- Largest
- 235B
- Runs from
- 8GB
The Qwen3 lineup
- Qwen3 0.6B0.6B · ~0.48 GB Q4_K_M · needs ~1 GB
- Qwen3 1.7B1.7B · ~1.28 GB Q4_K_M · needs ~2 GB
- Qwen3 4B4B · ~2.5 GB Q4_K_M · needs ~4 GB
- Qwen3 8B8B · ~5.03 GB Q4_K_M · needs ~7 GB
- Qwen3 14B14B · ~9 GB Q4_K_M · needs ~11 GB
- Qwen3 30B-A3B30.5B MoE · ~18.6 GB Q4_K_M · needs ~21 GB · Elo 1383
- Qwen3 32B32B · ~19.8 GB Q4_K_M · needs ~22 GB · Elo 1347
- Qwen3 235B A22B235B MoE · ~132.39 GB Q4_K_M · needs ~140 GB
"Needs" is the sourced minimum memory for Q4_K_M with a small context. Larger context needs more.
Which Qwen3 fits your memory
Largest that fits: Qwen3 32B (32B), best case on Nvidia GeForce RTX 4090 (24GB). Comfortable up to Qwen3 14B (14B).
Largest that fits: Qwen3 235B A22B (235B), best case on Apple M3 Ultra (256GB).
Best case means the most capable device at that size (usually a discrete GPU). A Mac at the same size sits roughly one rung lower; see the per-size breakdown on each memory budget page.
FAQ
Which Qwen3 size should I run locally?
Pick the largest size your memory allows. On 8GB (best case) up to Qwen3 4B; On 16GB (best case) up to Qwen3 14B; On 24GB (best case) up to Qwen3 32B; On 32GB (best case) up to Qwen3 32B; On 48GB (best case) up to Qwen3 32B; On 64GB (best case) up to Qwen3 32B; On 128GB (best case) up to Qwen3 32B; On 256GB (best case) up to Qwen3 235B A22B. Smaller sizes run faster and leave headroom for context.
What is the smallest Qwen3 model?
Qwen3 0.6B at 0.6B parameters, about 0.48 GB on disk at Q4_K_M and roughly 1 GB of memory to run. It is the one to use on phones and 8 GB machines.
What is the largest Qwen3 model and what does it need?
Qwen3 235B A22B at 235B (mixture of experts), about 132.39 GB at Q4_K_M and roughly 140 GB of memory. It fits a high-memory desktop GPU or Mac.
Understand the numbers
Short guides to the ideas behind Qwen3's memory and quant figures.
Sources
- apxml.com
- github.com
- gorilla.cs.berkeley.edu
- huggingface.co/bartowski/Qwen_Qwen3-0.6B-GGUF
- huggingface.co/bartowski/Qwen_Qwen3-1.7B-GGUF
- huggingface.co/Qwen/Qwen3-14B-GGUF
- huggingface.co/Qwen/Qwen3-30B-A3B-GGUF
- huggingface.co/Qwen/Qwen3-4B-GGUF
- huggingface.co/Qwen/Qwen3-8B-GGUF
- huggingface.co/unsloth
- lmarena.ai
- ollama.com/library/qwen3
- ollama.com/library/qwen3:0.6b
- ollama.com/library/qwen3:1.7b
- ollama.com/library/qwen3:4b
- qwenlm.github.io
Memory figures are estimates at Q4_K_M. See methodology.