Qwen3.5-35B-A3B requirements — can you run it?
~34.7B parameters · 23 quantizations measured · updated 2026-03-05 · source: unsloth/Qwen3.5-35B-A3B-GGUF
# llama.cpp (auto-downloads the GGUF) llama-cli -hf unsloth/Qwen3.5-35B-A3B-GGUF:UD-IQ2_M # Ollama (pulls straight from Hugging Face) ollama run hf.co/unsloth/Qwen3.5-35B-A3B-GGUF:UD-IQ2_M # LM Studio: search "unsloth/Qwen3.5-35B-A3B-GGUF" in the model browser
Commands load the exact quant measured on this page.
Every quantization: real file size vs what you actually need
| Quant | File (measured) | VRAM/RAM needed (8K ctx) | Runs on (first fit) |
|---|---|---|---|
| UD-IQ2_XXS | 10.7 GB | 12.3 GB | RTX 4060 Ti 16GB |
| UD-IQ2_M | 11.4 GB | 13.1 GB | RTX 4060 Ti 16GB |
| UD-Q2_K_XL | 12.2 GB | 13.8 GB | RTX 4060 Ti 16GB |
| UD-IQ3_XXS | 13.1 GB | 14.8 GB | RTX 3090 24GB |
| UD-IQ3_S | 13.6 GB | 15.3 GB | RTX 3090 24GB |
| Q3_K_S | 15.3 GB | 17.2 GB | RTX 3090 24GB |
| Q3_K_M | 16.4 GB | 18.3 GB | RTX 3090 24GB |
| UD-Q3_K_XL | 16.6 GB | 18.6 GB | RTX 3090 24GB |
| UD-IQ4_XS | 17.5 GB | 19.6 GB | RTX 3090 24GB |
| UD-IQ4_NL | 17.8 GB | 19.9 GB | RTX 3090 24GB |
| UD-Q4_K_L | 20.2 GB | 22.5 GB | RTX 5090 32GB |
| Q4_K_S | 20.7 GB | 23.0 GB | RTX 5090 32GB |
| Q4_K_M | 22.0 GB | 24.4 GB | RTX 5090 32GB |
| UD-Q4_K_XL | 22.2 GB | 24.7 GB | RTX 5090 32GB |
| Q5_K_S | 24.8 GB | 27.5 GB | RTX 5090 32GB |
| Q5_K_M | 26.2 GB | 29.0 GB | RTX 5090 32GB |
| UD-Q5_K_XL | 26.4 GB | 29.2 GB | RTX 5090 32GB |
| UD-Q6_K_S | 28.5 GB | 31.5 GB | RTX PRO 6000 96GB |
| Q6_K | 28.9 GB | 31.8 GB | RTX PRO 6000 96GB |
| UD-Q6_K_XL | 32.1 GB | 35.3 GB | RTX PRO 6000 96GB |
| Q8_0 | 36.9 GB | 40.5 GB | RTX PRO 6000 96GB |
| UD-Q8_K_XL | 48.7 GB | 53.3 GB | RTX PRO 6000 96GB |
| BF16 | 69.4 GB | 75.6 GB | RTX PRO 6000 96GB |
KV cache computed exactly from the model config (GQA formula, 8K context). MoE models keep all experts in memory — total size counts, not just active params. Longer context grows the KV cache — use the selector above.
Can I run Qwen3.5-35B-A3B on my GPU? (UD-IQ2_M recommended quant)
| Hardware | Memory | Verdict | Est. speed |
|---|---|---|---|
| RTX 3060 12GB | 12 GB | no | |
| RTX 4060 Ti 16GB | 16 GB | RUNS | ~125 tok/s |
| RTX 3090 24GB | 24 GB | RUNS | ~407 tok/s |
| RTX 4090 24GB | 24 GB | RUNS | ~439 tok/s |
| RTX 5090 32GB | 32 GB | RUNS | ~780 tok/s |
| RTX PRO 6000 96GB | 96 GB | RUNS | ~780 tok/s |
| Mac 16GB unified | 16 GB | RUNS | ~44 tok/s |
| Mac 32GB unified | 32 GB | RUNS | ~65 tok/s |
| Mac 64GB unified | 64 GB | RUNS | ~119 tok/s |
| Mac 128GB unified | 128 GB | RUNS | ~174 tok/s |
| Mac 256GB unified | 256 GB | RUNS | ~238 tok/s |
| Mac 512GB unified | 512 GB | RUNS | ~357 tok/s |
| 32GB system RAM (CPU) | 32 GB | RUNS | ~22 tok/s |
| 64GB system RAM (CPU) | 64 GB | RUNS | ~22 tok/s |
| 128GB system RAM (CPU) | 128 GB | RUNS | ~35 tok/s |
| 256GB system RAM (CPU) | 256 GB | RUNS | ~35 tok/s |
| 1TB server RAM (CPU) | 1024 GB | RUNS | ~87 tok/s |
| 2TB server RAM (CPU) | 2048 GB | RUNS | ~87 tok/s |
Speed = decode tok/s estimated from memory bandwidth ÷ measured file size (× MoE active share). Real numbers vary ±30% by runtime and settings.
How much VRAM does Qwen3.5-35B-A3B need?
The honest answer: 13.1 GB at the UD-IQ2_M quant with 8K context — that is the measured
file size (11.4 GB) plus KV cache and runtime overhead. The smallest published build needs
12.3 GB; the lossless one needs 75.6 GB.
Cheapest hardware that runs it: New RTX 4060 Ti 16GB (~$420, used market) — see which GPU to buy.
Can I run it with CPU offload?
Yes, if combined RAM+VRAM ≥ file size — but generation speed is limited by memory bandwidth. Expect roughly: DDR5 dual channel ~10-30 tok/s for small MoE actives, single digits for big ones, 0.1-1 tok/s when experts page from disk. CPU-only is fine for batch jobs, painful for chat.
Why do other sites show different numbers?
Most pages compute weights from a formula (params × bits/8). We use the actual file sizes published in GGUF repos, which include the embedding table, unquantized tensors, and container overhead — that is why our numbers can differ from a naive calculation.
FAQ
How much VRAM does Qwen3.5-35B-A3B need?
The measured answer: 13.1 GB at the UD-IQ2_M quant with 8K context. The smallest published build needs 12.3 GB; the lossless (F16/BF16) build needs 75.6 GB. These are real GGUF file sizes from Hugging Face, not formula estimates.
Can I run Qwen3.5-35B-A3B on a 24GB GPU (RTX 3090/4090)?
Yes — at UD-IQ2_M (13.1 GB). Quants that fit 24GB: UD-IQ2_XXS, UD-IQ2_M, UD-Q2_K_XL, UD-IQ3_XXS, UD-IQ3_S, Q3_K_S, Q3_K_M, UD-Q3_K_XL, UD-IQ4_XS, UD-IQ4_NL, UD-Q4_K_L, Q4_K_S.
How fast will Qwen3.5-35B-A3B run?
Decode speed is memory-bandwidth-bound. At UD-IQ2_M: roughly 439 tok/s on an RTX 4090, 407 tok/s on an RTX 3090, 174 tok/s on an M-series Mac with 128GB, and 21.8 tok/s on CPU with dual-channel DDR4 — estimates from measured file size, MoE active-parameter share and memory bandwidth.
Can I run Qwen3.5-35B-A3B on CPU with system RAM?
Yes, if your usable RAM is at least the file size. At UD-IQ2_M you need ~13.1 GB of RAM. CPU generation is memory-bandwidth-bound: expect single-digit tok/s for large MoE models, 10-30 tok/s for small active params, and below 1 tok/s when experts page from disk.
Why do VRAM numbers for Qwen3.5-35B-A3B differ between sites?
Most sites compute weights from a formula (params x bits/8). ModelFit uses the actual GGUF file sizes published on Hugging Face, which include the embedding table, unquantized tensors, and container overhead — so our numbers reflect what you really download and load.