---
title: Qwen3.5-397B-A17B-MLX-Q9
canonical_url: "https://www.modelscope.ai/models/inferencerlabs/Qwen3.5-397B-A17B-MLX-Q9"
md_url: "https://www.modelscope.ai/models/inferencerlabs/Qwen3.5-397B-A17B-MLX-Q9.md"
repository: inferencerlabs/Qwen3.5-397B-A17B-MLX-Q9
last_updated: 2026-06-28
pipeline_tag: image-text-to-text
tasks:
  - image-text-to-text
model_type:
  - qwen3_5_moe
architectures:
  - Qwen3_5MoeForConditionalGeneration
base_model:
  - Qwen/Qwen3.5-397B-A17B
base_model_relation: quantized
parameters: 124.3B
tensor_type:
  - U32
  - BF16
  - F32
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
language:
  - en
downloads: 233
stars: 0
tags:
  - quantized
  - mlx
---

# Qwen3.5-397B-A17B-MLX-Q9

> Qwen3.5-397B-A17B-MLX-Q9 - An open-source model by inferencerlabs on ModelScope. Qwen3.5-397B-A17B See Qwen3.5-397B-A17B in action: demonstration video

inferencerlabs/Qwen3.5-397B-A17B-MLX-Q9 is a 124.3B-parameter image-text-to-text model on ModelScope. derived from Qwen/Qwen3.5-397B-A17B.

- **Repository**: inferencerlabs/Qwen3.5-397B-A17B-MLX-Q9
- **Tasks**: image-text-to-text
- **Parameters**: 124.3B
- **Base model**: Qwen/Qwen3.5-397B-A17B
- **Tags**: quantized, mlx
- **Downloads**: 233
- **Stars**: 0
- **Last updated**: 2026-06-28

Source: https://www.modelscope.ai/models/inferencerlabs/Qwen3.5-397B-A17B-MLX-Q9

---

# Qwen3.5-397B-A17B
See Qwen3.5-397B-A17B in action: [demonstration video](https://youtu.be/tzF8jv3VGAg)

![Screenshot](https://cdn-uploads.huggingface.co/production/uploads/688479d616f1ec82fa645019/vRWau_2T-7GV9hE3w4Q2T.jpeg)

#### Tested on a M3 Ultra 512GB RAM using [Inferencer app](https://inferencer.com)
- Text inference: ~25.2 tokens/s @ 1000 tokens ~417 GiB
- Vision inference ~22.4 tokens/s @ 1000 tokens ~417 GiB
- Batched inference ~36 total tokens/s @ 2 inferences

*Q9-bit quant typically achieves near lossless accuracy in our coding test*
| Quantization | Perplexity | Token Accuracy | Missed Divergence |
|:------------:|:----------:|:--------------:|:-----------------:|
| **q3.5**     | 168.0      | 43.45%         | 72.57%            |
| **q4.5**     | 1.33593    | 91.65%         | 27.61%            |
| **q4.8**     | 1.28125    | 93.75%         | 21.15%            |
| **q5.5**     | 1.23437    | 95.05%         | 17.28%            |
| **q6.5**     | 1.21875    | 96.95%         | 12.03%            |
| **q8.5**     | 1.21093    | 97.55%         | 10.50%            |
| **q9**       | 1.21093    | 97.55%         | 10.50%            |
| **Base**     | 1.20312    | 100.0%         | 0.000%            |

- Perplexity: Measures the confidence for predicting base tokens (lower is better)
- Token Accuracy: The percentage of correctly generated base tokens
- Missed Divergence: Measures severity of misses; how much the token was missed by

##### Quantized with a modified version of [MLX](https://github.com/ml-explore/mlx)
##### For more details see [demonstration video](https://youtu.be/tzF8jv3VGAg) or visit [Qwen3.5-397B-A17B](https://huggingface.co/Qwen/Qwen3.5-397B-A17B).

## Disclaimer

We are not the creator, originator, or owner of any model listed. Each model is created and provided by third parties. Models may not always be accurate or contextually appropriate. You are responsible for verifying the information before making important decisions. We are not liable for any damages, losses, or issues arising from its use, including data loss or inaccuracies in AI-generated content.