Qwen 3.5 397B-A17B
MoEThinkingVisionAlibaba · 397B (17B active) · Mixture of Experts
Largest multimodal Qwen 3.5 MoE
1.4M
1.3K
2026-02
256K tokens
Use cases
Your hardware
16 GB Reference Desktop
Quantization options
What is quantization?| Quant | Bits | VRAM | Disk | Quality | Status |
|---|---|---|---|---|---|
| Q2_K | 2 | 127.6 GB | 121.3 GB | low | Too heavy · ~1 t/s |
| Q3_K_M | 3 | 178.4 GB | 169.8 GB | moderate | Too heavy · ~1 t/s |
| Q4_K_M | 4 | 203.9 GB | 194.1 GB | good | Too heavy · ~1 t/s |
| Q5_K_M | 5 | 254.7 GB | 242.6 GB | good | Too heavy · ~1 t/s |
| Q6_K | 6 | 305.5 GB | 291.2 GB | excellent | Too heavy · ~1 t/s |
| Q8_0 | 8 | 407.2 GB | 388.2 GB | excellent | Too heavy · ~1 t/s |
| F16 | 16 | 813.9 GB | 776.4 GB | lossless | Too heavy · ~1 t/s |
About this model
Qwen 3.5 397B-A17B (Qwen3.5 397b) is a 397B sparse mixture-of-experts (MoE) model activating 17B active per token from Alibaba in the Qwen family, designed primarily for chat, vision, reasoning, code workflows. Native reasoning/thinking tokens expand output length per query, so reserve extra KV cache headroom. Supports up to 256K context tokens; start at 4K–8K with Q8_0 KV cache on tight GPUs.
Hardware planning baselines start at 221.8 GB minimum memory and 369.7 GB recommended memory. At Q4_K_M, weights plus baseline runtime overhead use roughly 203.9 GB of RAM/VRAM and 194.1 GB of disk space. At 397B, Qwen 3.5 397B-A17B belongs firmly in the workstation and multi-GPU tier. Running it comfortably requires 369.7 GB of total memory across dual RTX GPUs, high-memory Apple Silicon (Mac Studio Ultra/Max), or MoE hybrid CPU-RAM offloading—mobile devices and standard 16 GB laptops are too constrained for practical speeds.
Deploy Qwen 3.5 397B-A17B locally via Ollama (`qwen3.5:397b-cloud`), Ollama with Flash Attention enabled. Under Apache License 2.0, review official weight cards before commercial redistribution.
Highlights
- Local Fit Baseline221.8 GB minimum and 369.7 GB recommended RAM/VRAM.
- Primary QuantQ4_K_M targets 203.9 GB memory and 194.1 GB disk.
- Context & Capabilities256K max context · Sparse MoE · Reasoning.
System requirements by device
Memory planning targets 221.8 GB minimum RAM/VRAM and 369.7 GB recommended memory. The baseline quant is Q4_K_M, estimated at 203.9 GB memory and 194.1 GB disk. Actual runtime usage shifts with context length, KV cache precision, and OS background overhead.
| Device | Notes |
|---|---|
| Mac (Apple silicon) | Requires a 48 GB–192 GB Mac Studio or MacBook Pro Max/Ultra with enough unified wired budget for 203.9 GB. |
| Windows / Linux GPU | Use dual/multi-GPU CUDA splitting (--tensor-split) or MoE CPU expert offloading (-ot) paired with 32–128 GB DDR5 system RAM. |
| Phones & Compact Edge Boards | Not practical on phones (221.8 GB min RAM exceeds Android LMK and iOS Jetsam limits). Pick a 2B–4B model on mobile. |
Start with a short session and leave memory for the operating system. A file fitting on disk does not guarantee that it will load in memory.
Estimated memory by quantization
These are rough estimates, not tested minimum requirements. Check the actual file and runtime documentation; context, offloading and operating-system overhead can change the memory needed.
| Quant Tier | Estimated memory | Disk Size | Quality |
|---|---|---|---|
| Q2_K | 127.6 GB | 121.3 GB | low |
| Q3_K_M | 178.4 GB | 169.8 GB | moderate |
| Q4_K_M | 203.9 GB | 194.1 GB | good |
| Q5_K_M | 254.7 GB | 242.6 GB | good |
| Q6_K | 305.5 GB | 291.2 GB | excellent |
| Q8_0 | 407.2 GB | 388.2 GB | excellent |
| F16 | 813.9 GB | 776.4 GB | lossless |
How to run Qwen 3.5 397B-A17B on your device
Start with a short prompt and the smallest quant that still gives the quality you need. A clean first run is more useful than loading a larger file with no memory headroom.
Run Qwen 3.5 397B-A17B on Multi-GPU or High-RAM Desktop
- Plan for 221.8 GB minimum combined memory and 369.7 GB recommended memory for Q4_K_M.
- Use llama.cpp `-ot ".ffn_.*_exps.=CPU"` or `--tensor-split` across multiple GPUs so attention and KV cache remain in fast VRAM.
- Enable Flash Attention (`-fa on`) and Q8_0 KV cache (`-ctk q8_0 -ctv q8_0`) to prevent context buffers from exceeding GPU headroom.
- Avoid attempting to run workstation-scale weights on smartphones or 8 GB laptops; use a 2B–8B checkpoint on portable devices.
Run Qwen 3.5 397B-A17B on High-Memory Apple Silicon Mac
- Verify you are on a 48 GB–192 GB Mac Studio or MacBook Pro Max/Ultra capable of holding 203.9 GB in wired unified memory.
- Launch via `llama-server` or LM Studio with MLX/Metal enabled and set initial context conservatively to 4K–8K.
- Monitor Activity Monitor Memory Pressure to confirm zero SSD swap thrashing during token generation.
- Step down to a lighter quantization (e.g., Q3_K_M or IQ4_XS) if total memory utilization exceeds 75%.
Qwen 3.5 397B-A17B vs alternatives
| Model | Params | Context | Min RAM | License |
|---|---|---|---|---|
| Qwen 3.5 397B-A17B | 397B (17B active) | 256K | 221.8 GB | Apache 2.0 |
| Qwen 3 Coder 480B | 480B (35B active) | 256K | 268.2 GB | Apache 2.0 |
| Qwen 3 235B-A22B | 235B (22B active) | 128K | 131.3 GB | Apache 2.0 |
| Qwen 3 VL 235B-A22B | 235B (22B active) | 256K | 131.3 GB | Apache 2.0 |
| Qwen 3.8 Flash 128B | 125B (6B active) | 128K | 32 GB | Apache 2.0 |
| Qwen 3.5 122B-A10B | 122B (10B active) | 256K | 68.2 GB | Apache 2.0 |
Use this table to choose the smallest nearby model that still matches the job. A related model with lower RAM can be the better local choice when the current page fits only at the edge of available memory.
Frequently asked questions
- What is Qwen 3.5 397B-A17B?
- Qwen 3.5 397B-A17B is a 397B sparse mixture-of-experts model (17B active active per token) from Alibaba, supporting up to 256K tokens with 221.8 GB minimum RAM and 369.7 GB recommended RAM.
- What is the best way to run Qwen3.5 397b locally?
- Choose the smallest quant that matches the task, test one short prompt, then compare RAM, context, and nearby models before making Qwen 3.5 397B-A17B your default local model.
- How much VRAM or RAM does Qwen 3.5 397B-A17B require?
- At Q4_K_M, Qwen 3.5 397B-A17B occupies roughly 203.9 GB of active memory and 194.1 GB on disk. Add 2–6 GB of spare headroom for KV cache buffers, runtime scratch space, and OS apps (369.7 GB recommended).
- What is the best way to run Qwen 3.5 397B-A17B locally?
- Use `ollama run qwen3.5:397b-cloud` for a quick CLI/API setup, or load Q4_K_M in LM Studio or llama.cpp with Flash Attention enabled when you want explicit KV cache and layer control.
- Can Qwen 3.5 397B-A17B run on a consumer laptop or smartphone?
- No smartphone or 8–16 GB laptop can hold Qwen 3.5 397B-A17B cleanly in local memory. It targets multi-GPU NVIDIA workstations, high-RAM Mac Studios, or MoE CPU-RAM hybrid rigs with at least 221.8 GB total memory.
- How does Qwen 3.5 397B-A17B compare to Qwen 3 Coder 480B?
- Qwen 3 Coder 480B (480B, 268.2 GB min RAM) sits closest in size. Compare active parameters, context ceiling (256K vs 256K), and license terms (Apache 2.0) before switching.