MODELS

Qwen: Qwen3.6 35B A3B

by Qwen

modellead-sourceopenrouter-modelssource:github.comqwenqwen3.6multimodalopen-weightmoehugging-facegithubopenrouter

Overview

Open-weight multimodal Qwen model with 35B total parameters and 3B active parameters per token.

Details

Qwen3.6-35B-A3B is described by OpenRouter as an open-weight multimodal model from Alibaba Cloud with 35 billion total parameters and 3 billion active parameters per token, using a hybrid sparse mixture-of-experts architecture. The official Hugging Face model card is cited as providing model weights and configuration files compatible with Transformers, vLLM, SGLang, and KTransformers, along with serving commands for the Qwen/Qwen3.6-35B-A3B model ID. The QwenLM GitHub repository describes Qwen3.6 as a large language model series from the Qwen team at Alibaba Group and states that this model became available on Hugging Face Hub and ModelScope on 2026-04-16.

When to Use

Use when you need to evaluate an open-weight multimodal Qwen model with a sparse mixture-of-experts design. Use when your serving stack is based on Transformers, vLLM, SGLang, or KTransformers and you want a model card that includes compatible weights/configuration and serving commands. Use when comparing Qwen3.6 models available through Hugging Face, ModelScope, GitHub, or OpenRouter discovery listings.

Getting Started

Review the Qwen/Qwen3.6-35B-A3B Hugging Face model card for weights, configuration files, compatibility notes, and serving commands.
Check the QwenLM/Qwen3.6 GitHub repository for Qwen3.6 series information and release context.
If using OpenRouter, review the OpenRouter model page for provider-side availability before routing requests through that catalog.
Run a small compatibility test in your intended serving framework before production use.

Key Features

•35B total parameters with 3B active parameters per token, per the OpenRouter listing.
•Open-weight multimodal model, per the OpenRouter listing.
•Hybrid sparse mixture-of-experts architecture, per the OpenRouter listing.
•Hugging Face-hosted weights and configuration files compatible with Transformers, vLLM, SGLang, and KTransformers.
•Part of the Qwen3.6 large language model series from the Qwen team at Alibaba Group.

Capabilities

•multimodal-model
•open-weight-model
•mixture-of-experts
•transformers-compatible
•vllm-compatible
•sglang-compatible
•ktransformers-compatible

Last updated May 31, 2026