Qwen3.6-27B

Updated

Tools

Thinking

Embedding

Vision

Reasoning

Code

Multilingual

Run

Run Qwen3.6-27B locally in Atomic Chat — strong reasoning and coding, offline and private. No cloud, no usage limits. Download free.

huggingface-cli download Qwen/Qwen3.6-27B

from transformers import AutoModel
model = AutoModel.from_pretrained("Qwen/Qwen3.6-27B")

More models

View all

Name	Size / Usage	Context	Input
Qwen3.6-35B-A3B		256K	Text, Image
Qwen3-8B	Reasoning, coding, agentic chat	128K	Text
Qwen2.5-7B-Instruct	General chat, coding, reasoning	128K	Text
Qwen2.5-32B-Instruct	General chat, coding, reasoning	128K	Text
Qwen2.5-14B-Instruct	General chat, coding, reasoning	128K	Text
Qwen2.5-Coder-7B-Instruct	Code generation, code reasoning	128K	Text
Qwen2.5-Coder-32B-Instruct	Code generation, fixing, agents	128K	Text
Qwen3-235B-A22B	Reasoning, coding, agentic tasks	128K	Text
Qwen3-30B-A3B-Instruct-2507	General chat, reasoning, agents	256K	Text
Qwen2.5-72B-Instruct	General chat, coding, reasoning	128K	Text
Qwen3-4B-Thinking-2507	Reasoning, math, coding	256K	Text
WebWorld-8B	Web agents, multimodal reasoning	40K	Text, Image
MiniCPM-V 4.6	5213 GB	421K	Text, Image
anima	0.4B	31K	Text
Qwen3-Coder-30B-A3B-Instruct		256K	Text
Qwen3-30B-A3B		128K	Text
Qwen3-14B		128K	Text
Qwen3-32B		128K	Text

At a glance

License: Apache 2.0
Context length: 256K tokens
Languages: Multilingual
Minimum hardware: ~16 GB VRAM
Strengths: reasoning, coding and on-device inference

Overview

Qwen3.6-27B is a 27.8B-parameter dense model from Qwen, the model team at Alibaba Cloud.

License: Apache 2.0
Context length: 256K tokens
Languages: Multilingual
Minimum hardware: ~16 GB VRAM
Strengths: reasoning, coding and on-device inference

Highlights

Agentic coding — handles frontend workflows and repository-level reasoning with greater fluency.
Thinking preservation — retains reasoning context from earlier messages to streamline iterative work.
Vision + document understanding — reads screenshots, scanned pages, and long video.
256K context — dense 27.8B model for long-document and code work.

What it is good at

The capability set covers reasoning, code, vision, tool calling, and embeddings, so a single local copy handles a wide range of work:

Agentic coding — Qwen reports Qwen3.6-27B scoring 77.2 on SWE-bench Verified and 59.3 on Terminal-Bench 2.0, ahead of the older Qwen3.5-397B-A17B flagship, which makes it suited to repository-level edits and frontend work.
Vision and document understanding — the built-in vision encoder reads screenshots, scanned pages with OCR, and long video, so you can ask questions about an image or a PDF without an external service.
Tool calling with thinking — the model supports structured tool calls and a thinking mode, letting it plan multi-step tasks and invoke functions inside an agent loop.

Running it locally

At 27.8B parameters, a 16 GB GPU such as an RTX 4080 or 16 GB of Apple unified memory runs the Q4_K_M quantization at around 16.8 GB. A 24 GB card like the RTX 3090 or 4090 gives headroom for Q6_K and longer slices of the 262,144-token context. Pull the original weights from Hugging Face:

huggingface-cli download Qwen/Qwen3.6-27B

From there you can serve it with Transformers or vLLM, load a GGUF build in LM Studio, or use the one-click download in Atomic Chat to get it running without touching the terminal.

License

Qwen3.6-27B is released under the Apache-2.0 license. That permits commercial use, modification, redistribution, and private deployment with no royalties, so you can build products on top of the model and run it on your own hardware without a usage fee.

Desktop

macOS

(M1 or better)

Download

Windows

(x64)

Download

Linux

(x86_64)

Download

Frequently asked questions

Qwen3.6-27B is a 27.8B-parameter dense, multimodal language model from Qwen (Alibaba Cloud). It handles text, images, and video, supports reasoning and tool calling, and carries a 262,144-token context window. The weights are open under Apache-2.0, so it can run locally in Atomic Chat without an API.

The Q4_K_M quantization needs about 16.8 GB, so a 16 GB GPU like an RTX 4080 or 16 GB of Apple unified memory is the practical floor. A 24 GB card such as an RTX 3090 or RTX 4090 gives room for the higher-quality Q6_K build and longer context. Lower quants like Q3_K_M drop the requirement to around 12 GB if memory is tight.

Yes. Qwen3.6-27B is published under the Apache-2.0 license, which allows free use including commercial deployment, with no royalties or usage fees. You can download the weights from Hugging Face and run them on your own hardware at no cost.

Yes. Once the weights are downloaded, the model runs entirely on your machine with no network connection. In Atomic Chat your prompts, code, and files never leave your device, which suits private or air-gapped work.

Its strongest area is agentic coding, where Qwen reports 77.2 on SWE-bench Verified, ahead of the larger Qwen3.5-397B-A17B model. It is also natively multimodal, reading images, OCR documents, and long video, and it supports tool calling plus a thinking mode for multi-step agent tasks.