What are the hardware requirements for Qwen3.6 35B A3B MLX 4bit MTP?

Qwen3.6 35B A3B MLX 4bit MTP requires approximately 20.5 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Qwen3.6 35B A3B MLX 4bit MTP and how large is it?

Qwen3.6 35B A3B MLX 4bit MTP is developed by Vontra, a model with 35b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Qwen3.6 35B A3B MLX 4bit MTP perform on standard benchmarks?

Qwen3.6 35B A3B MLX 4bit MTP has the following published scores: LMSYS ELO 1395. Compare against reference models on this page or on the LLM Explorer leaderboards.

Qwen3.6 35B A3B MLX 4bit MTP by Vontra — VRAM 20.5GB

Name: Qwen3.6 35B A3B MLX 4bit MTP
Author: Vontra

Qwen3.6 35B A3B MLX 4bit MTP is an open-source language model by Vontra. Features: 35b LLM, VRAM: 20.5GB, License: apache-2.0, Quantized, LLM Explorer Score: 0.45, ELO: 1395.

4-bit 4bit Base model:quantized:qwen/qwen... Base model:qwen/qwen3.6-35b-a3... Conversational Image-text-to-text Mixture-of-experts Mlx Mlx-vlm Mtp Quantized Qwen Qwen3.6 Qwen3 5 moe Region:us Safetensors Sharded Speculative-decoding Tensorflow

Model Card on HF 🤗: https://huggingface.co/TensorFold/Qwen3.6-35B-A3B-MLX-4bit-MTP

Qwen3.6 35B A3B MLX 4bit MTP Benchmarks

LMSys ELO 1395 vs 1272 (so35) 9.7%

^nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

What is the LLM Explorer Rank (Score)

Qwen3.6 35B A3B MLX 4bit MTP Parameters and Internals

LLM Name	Qwen3.6 35B A3B MLX 4bit MTP
Repository 🤗	https://huggingface.co/TensorFold/Qwen3.6-35B-A3B-MLX-4bit-MTP
Base Model(s)	Qwen/Qwen3.6-35B-A3B Qwen/Qwen3.6-35B-A3B
Model Size	35b
Required VRAM	20.5 GB
Updated	2026-10-02
Maintainer	Vontra
Model Type	qwen3_5_moe
Model Files	5.3 GB: 1-of-4 5.4 GB: 2-of-4 5.4 GB: 3-of-4 4.4 GB: 4-of-4 0.5 GB
Quantization Type	4bit
Model Architecture	Qwen3_5MoeForConditionalGeneration
License	apache-2.0
Model Max Length	262144
Transformers Version	4.57.1
Tokenizer Class	TokenizersBackend
Padding Token	<\|endoftext\|>
Errors	replace

Rank the Qwen3.6 35B A3B MLX 4bit MTP capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to Qwen3.6 35B A3B MLX 4bit MTP

Best Alternatives	Context / RAM	Downloads	Likes
Qwen3.6 35B A3B MLX 4bit	0K / 20.5 GB	513193	6
Qwen3.6 35B A3B MLX 6bit	0K / 28.9 GB	478592	2
Qwen3.6 35B A3B MLX 8bit	0K / 37.8 GB	473537	1
Qwen3.6 35B A3B UD MLX 4bit	0K / 21.6 GB	15720	112
Qwen3.6 35B A3B MLX 8bit	0K / 37.8 GB	3862	41
Qwen3.6 35B A3B OQ4e Fp16 Mtp	0K / 22.4 GB	187	1
Qwen3.6 35B A3B 2Bit GSQ	0K / 13.5 GB	2119	7
Qwen3.6 35B A3B 6bit	0K / 28.9 GB	2742	8
Qwen3.6 35B A3B UD MLX 3bit	0K / 17.3 GB	2292	13
Qwen3.6 35B A3B 5bit	0K / 24.7 GB	1628	5

Note: green Score (e.g. "73.2") means that the model is better than TensorFold/Qwen3.6-35B-A3B-MLX-4bit-MTP.