image-text-to-texttransformersapache-2.0

unsloth/Qwen3.6-35B-A3B-GGUF

huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF

1,600Likes
1,289,328Downloads
2026-04-20Updated
transformersggufunslothqwenqwen3_5_moeimage-text-to-textbase_model:Qwen/Qwen3.6-35B-A3Bbase_model:quantized:Qwen/Qwen3.6-35B-A3Blicense:apache-2.0endpoints_compatibleregion:usimatrixconversationaldeploy:azure

Model card

Read our How to Run Qwen3.6 Guide!

See Unsloth Dynamic 2.0 GGUFs for our quantization benchmarks.

NEW: Developer Role Support so Qwen3.6 can work in Codex, OpenCode and more!

Qwen3.6 can now be run and fine-tuned in Unsloth Studio. Read our guide.

Tool calling improvements: Makes parsing nested objects to make tool calling succeed more.

Example of Qwen3.6 (4-bit GGUF) running in Unsloth Studio with tool-calling:

Qwen3.6-35B-A3B

![Qwen Chat](https://chat.qwen.ai)

Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience.

Qwen3.6 Highlights

This release delivers substantial upgrades, particularly in

!Benchmark Results

For more details, please refer to our blog post Qwen3.6-35B-A3B.

Model Overview

Benchmark Results

Language

· Qwen3.5-27B · Gemma4-31B · Qwen3.5-35BA3B · Gemma4-26BA4B · Qwen3.6-35BA3B ·

Coding Agent ·

SWE-bench Verified · 75.0 · 52.0 · 70.0 · 17.4 · 73.4 ·

SWE-bench Multilingual · 69.3 · 51.7 · 60.3 · 17.3 · 67.2 ·

SWE-bench Pro · 51.2 · 35.7 · 44.6 · 13.8 · 49.5 ·

Terminal-Bench 2.0 · 41.6 · 42.9 · 40.5 · 34.2 · 51.5 ·

Claw-Eval Avg · 64.3 · 48.5 · 65.4 · 58.8 · 68.7 ·

Claw-Eval Pass^3 · 46.2 · 25.0 · 51.0 · 28.0 · 50.0 ·

SkillsBench Avg5 · 27.2 · 23.6 · 4.4 · 12.3 · 28.7 ·

QwenClawBench · 52.2 · 41.7 · 47.7 · 38.7 · 52.6 ·

NL2Repo · 27.3 · 15.5 · 20.5 · 11.6 · 29.4 ·

QwenWebBench · 1068 · 1197 · 978 · 1178 · 1397 ·

General Agent ·

TAU3-Bench · 68.4 · 67.5 · 68.9 · 59.0 · 67.2 ·

VITA-Bench · 41.8 · 43.0 · 29.1 · 36.9 · 35.6 ·

DeepPlanning · 22.6 · 24.0 · 22.8 · 16.2 · 25.9 ·

Tool Decathlon · 31.5 · 21.2 · 28.7 · 12.0 · 26.9 ·

MCPMark · 36.3 · 18.1 · 27.0 · 14.2 · 37.0 ·

MCP-Atlas · 68.4 · 57.2 · 62.4 · 50.0 · 62.8 ·

WideSearch · 66.4 · 35.2 · 59.1 · 38.3 · 60.1 ·

Knowledge ·

MMLU-Pro · 86.1 · 85.2 · 85.3 · 82.6 · 85.2 ·

MMLU-Redux · 93.2 · 93.7 · 93.3 · 92.7 · 93.3 ·

SuperGPQA · 65.6 · 65.7 · 63.4 · 61.4 · 64.7 ·

C-Eval · 90.5 · 82.6 · 90.2 · 82.5 · 90.0 ·

STEM & Reasoning ·

GPQA · 85.5 · 84.3 · 84.2 · 82.3 · 86.0 ·

HLE · 24.3 · 19.5 · 22.4 · 8.7 · 21.4 ·

LiveCodeBench v6 · 80.7 · 80.0 · 74.6 · 77.1 · 80.4 ·

HMMT Feb 25 · 92.0 · 88.7 · 89.0 · 91.7 · 90.7 ·

HMMT Nov 25 · 89.8 · 87.5 · 89.2 · 87.5 · 89.1 ·

HMMT Feb 26 · 84.3 · 77.2 · 78.7 · 79.0 · 83.6 ·

IMOAnswerBench · 79.9 · 74.5 · 76.8 · 74.3 · 78.9 ·

AIME26 · 92.6 · 89.2 · 91.0 · 88.3 · 92.7 ·

Vision Language

· Qwen3.5-27B · Claude-Sonnet-4.5 · Gemma4-31B · Gemma4-26BA4B · Qwen3.5-35B-A3B · Qwen3.6-35B-A3B ·

STEM and Puzzle ·

MMMU · 82.3 · 79.6 · 80.4 · 78.4 · 81.4 · 81.7 ·

MMMU-Pro · 75.0 · 68.4 · 76.9* · 73.8* · 75.1 · 75.3 ·

Mathvista(mini) · 87.8 · 79.8 · 79.3 · 79.4 · 86.2 · 86.4 ·

ZEROBench_sub · 36.2 · 26.3 · 26.0 · 26.3 · 34.1 · 34.4 ·

General VQA ·

RealWorldQA · 83.7 · 70.3 · 72.3 · 72.2 · 84.1 · 85.3 ·

MMBenchEN-DEV-v1.1 · 92.6 · 88.3 · 90.9 · 89.0 · 91.5 · 92.8 ·

SimpleVQA · 56.0 · 57.6 · 52.9 · 52.2 · 58.3 · 58.9 ·

HallusionBench · 70.0 · 59.9 · 67.4 · 66.1 · 67.9 · 69.8 ·

Text Recognition and Document Understanding ·

OmniDocBench1.5 · 88.9 · 85.8 · 80.1 · 74.4 · 89.3 · 89.9 ·

CharXiv(RQ) · 79.5 · 67.2 · 67.9 · 69.0 · 77.5 · 78.0 ·

CC-OCR · 81.0 · 68.1 · 75.7 · 74.5 · 80.7 · 81.9 ·

AI2D_TEST · 92.9 · 87.0 · 89.0 · 88.3 · 92.6 · 92.7 ·

Spatial Intelligence ·

RefCOCO(avg) · 90.9 · -- · -- · -- · 89.2 · 92.0 ·

ODInW13 · 41.1 · -- · -- · -- · 42.6 · 50.8 ·

EmbSpatialBench · 84.5 · 71.8 · -- · -- · 83.1 · 84.3 ·

RefSpatialBench · 67.7 · -- · -- · -- · 63.5 · 64.3 ·

Video Understanding ·

VideoMME(w sub.) · 87.0 · 81.1 · -- · -- · 86.6 · 86.6 ·

VideoMME(w/o sub.) · 82.8 · 75.3 · -- · -- · 82.5 · 82.5 ·

VideoMMMU · 82.3 · 77.6 · 81.6 · 76.0 · 80.4 · 83.7 ·

MLVU · 85.9 · 72.8 · -- · -- · 85.6 · 86.2 ·

MVBench · 74.6 · -- · -- · -- · 74.8 · 74.6 ·

LVBench · 73.6 · -- · -- · -- · 71.4 · 71.4 ·

Quickstart

For streamlined integration, we recommend using Qwen3.6 via APIs. Below is a guide to use Qwen3.6 via OpenAI-compatible API.

Serving Qwen3.6

Qwen3.6 can be served via APIs with popular inference frameworks. In the following, we show example commands to launch OpenAI-Compatible API servers for Qwen3.6 models.

SGLang

SGLang is a fast serving framework for large language models and vision language models. sglang>=0.5.10 is recommended for Qwen3.6, which can be installed using the following command in a fresh environment: ``shell uv pip install sglang[all] `` See its documentation for more details.

The following will create API endpoints at http://localhost:8000/v1:

- Standard Version: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs. ``shell python -m sglang.launch_server --model-path Qwen/Qwen3.6-35B-A3B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 ``

- Tool Use: To support tool use, you can use the following command. ``shell python -m sglang.launch_server --model-path Qwen/Qwen3.6-35B-A3B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --tool-call-parser qwen3_coder ``

- Multi-Token Prediction (MTP): The following command is recommended for MTP: ``shell python -m sglang.launch_server --model-path Qwen/Qwen3.6-35B-A3B --port 8000 --tp-size 8 --mem-fraction-static 0.8 --context-length 262144 --reasoning-parser qwen3 --speculative-algo NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 ``

For detailed deployment guide, see the SGLang Qwen3.5 Cookbook.

vLLM

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. vllm>=0.19.0 is recommended for Qwen3.6, which can be installed using the following command in a fresh environment: ``shell uv pip install vllm --torch-backend=auto `` See its documentation for more details.

The following will create API endpoints at http://localhost:8000/v1:

``shell vllm serve Qwen/Qwen3.6-35B-A3B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 ``

- Tool Call: To support tool use, you can use the following command. ``shell vllm serve Qwen/Qwen3.6-35B-A3B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder ``

``shell vllm serve Qwen/Qwen3.6-35B-A3B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}' ``

- Text-Only: The following command skips the vision encoder and multimodal profiling to free up memory for additional KV cache: ``shell vllm serve Qwen/Qwen3.6-35B-A3B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only ``

For detailed deployment guide, see the vLLM Qwen3.5 Recipe.

KTransformers

KTransformers is a flexible framework for experiencing cutting-edge LLM inference optimizations with CPU-GPU heterogeneous computing. For running Qwen3.6 with KTransformers, see the KTransformers Deployment Guide. #### Hugging Face Transformers

Hugging Face Transformers contains a _lightweight_ server which can be used for quick testing and moderate load deployment. The latest transformers is required for Qwen3.6: ``shell pip install "transformers[serving]" `` See its documentation for more details. Please also make sure torchvision and pillow are installed.

Then, run transformers serve to launch a server with API endpoints at http://localhost:8000/v1; it will place the model on accelerators if available: ``shell transformers serve Qwen/Qwen3.6-35B-A3B --port 8000 --continuous-batching ``

Using Qwen3.6 via the Chat Completions API

The chat completions API is accessible via standard HTTP requests or OpenAI SDKs. Here, we show examples using the OpenAI Python SDK.

Before starting, make sure it is installed and the API key and the API base URL is configured, e.g.: ```shell pip install -U openai

Set the following accordingly

export OPENAI_BASE_URL="http://localhost:8000/v1" export OPENAI_API_KEY="EMPTY" ```

Text-Only Input

```python from openai import OpenAI # Configured by environment variables client = OpenAI()

messages = [ {"role": "user", "content": "Type \"I love Qwen3.6\" backwards"}, ]

chat_response = client.chat.completions.create( model="Qwen/Qwen3.6-35B-A3B", messages=messages, max_tokens=81920, temperature=1.0, top_p=0.95, presence_penalty=1.5, extra_body={ "top_k": 20, }, ) print("Chat response:", chat_response) ```

Image Input

```python from openai import OpenAI # Configured by environment variables client = OpenAI()

messages = [ { "role": "user", "content": [ { "type": "image_url", "image_url": { "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/CI_Demo/mathv-1327.jpg" } }, { "type": "text", "text": "The centres of the four illustrated circles are in the corners of the square. The two big circles touch each other and also the two little circles. With which factor do you have to multiply the radii of the little circles to obtain the radius of the big circles?\nChoices:\n(A) $\\frac{2}{9}$\n(B) $\\sqrt{5}$\n(C) $0.8 \\cdot \\pi$\n(D) 2.5\n(E) $1+\\sqrt{2}$" } ] } ]

response = client.chat.completions.create( model="Qwen/Qwen3.6-35B-A3B", messages=messages, max_tokens=81920, temperature=1.0, top_p=0.95, presence_penalty=1.5, extra_body={ "top_k": 20, }, ) print("Chat response:", chat_response) ```

Video Input

```python from openai import OpenAI # Configured by environment variables client = OpenAI()

messages = [ { "role": "user", "content": [ { "type": "video_url", "video_url": { "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/video/N1cdUjctpG8.mp4" } }, { "type": "text", "text": "How many porcelain jars were discovered in the niches located in the primary chamber of the tomb?" } ] } ]

When vLLM is launched with --media-io-kwargs '{"video": {"num_frames": -1}}',

# video frame sampling can be configured via extra_body (e.g., by setting fps). # This feature is currently supported only in vLLM. # # By default, fps=2 and do_sample_frames=True. # With do_sample_frames=True, you can customize the fps value to set your desired video sampling rate. response = client.chat.completions.create( model="Qwen/Qwen3.6-35B-A3B", messages=messages, max_tokens=81920, temperature=1.0, top_p=0.95, presence_penalty=1.5, extra_body={ "top_k": 20, "mm_processor_kwargs": {"fps": 2, "do_sample_frames": True}, }, )

print("Chat response:", chat_response) ```

Instruct (or Non-Thinking) Mode

Qwen3.6 will think by default before response. You can obtain direct response from the model without thinking by configuring the API parameters. For example, ```python from openai import OpenAI # Configured by environment variables client = OpenAI()

messages = [ { "role": "user", "content": [ { "type": "image_url", "image_url": { "url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.6/demo/RealWorld/RealWorld-04.png" } }, { "type": "text", "text": "Where is this?" } ] } ]

chat_response = client.chat.completions.create( model="Qwen/Qwen3.6-35B-A3B", messages=messages, max_tokens=32768, temperature=0.7, top_p=0.8, presence_penalty=1.5, extra_body={ "top_k": 20, "chat_template_kwargs": {"enable_thinking": False}, }, ) print("Chat response:", chat_response) ```

Preserve Thinking

By default, only the thinking blocks generated in handling the latest user message is retained, resulting in a pattern commonly as interleaved thinking. Qwen3.6 has been additionally trained to preserve and leverage thinking traces from historical messages. You can enable this behavior by setting the preserve_thinking option: ```python from openai import OpenAI # Configured by environment variables client = OpenAI()

messages = [...]

chat_response = client.chat.completions.create( model="Qwen/Qwen3.6-35B-A3B", messages=messages, max_tokens=32768, temperature=0.7, top_p=0.8, presence_penalty=1.5, extra_body={ "top_k": 20, "chat_template_kwargs": {"preserve_thinking": True}, }, ) print("Chat response:", chat_response) ```

This capability is particularly beneficial for agent scenarios, where maintaining full reasoning context can enhance decision consistency and, in many cases, reduce overall token consumption by minimizing redundant reasoning. Additionally, it can improve KV cache utilization, optimizing inference efficiency in both thinking and non-thinking modes.

Agentic Usage

Qwen3.6 excels in tool calling capabilities.

Qwen-Agent

We recommend using Qwen-Agent to quickly build Agent applications with Qwen3.6.

To define the available tools, you can use the MCP configuration file, use the integrated tool of Qwen-Agent, or integrate other tools by yourself. ```python import os from qwen_agent.agents import Assistant

Define LLM

# Using Alibaba Cloud Model Studio llm_cfg = { # Use the OpenAI-compatible model service provided by DashScope: 'model': 'Qwen3.6-35B-A3B', 'model_type': 'qwenvl_oai', 'model_server': 'https://dashscope.aliyuncs.com/compatible-mode/v1', 'api_key': os.getenv('DASHSCOPE_API_KEY'),

'generate_cfg': { 'use_raw_api': True, # When using Dash Scope OAI API, pass the parameter of whether to enable thinking mode in this way 'extra_body': { 'enable_thinking': True, 'preserve_thinking': True, }, }, }

Using OpenAI-compatible API endpoint.

# functionality of the deployment frameworks and let Qwen-Agent automate the related operations. # # llm_cfg = { # # Use your own model service compatible with OpenAI API by vLLM/SGLang: # 'model': 'Qwen/Qwen3.6-35B-A3B', # 'model_type': 'qwenvl_oai', # 'model_server': 'http://localhost:8000/v1', # api_base # 'api_key': 'EMPTY', # # 'generate_cfg': { # 'use_raw_api': True, # # When using vLLM/SGLang OAI API, pass the parameter of whether to enable thinking mode in this way # 'extra_body': { # 'chat_template_kwargs': {'enable_thinking': True, 'preserve_thinking': True} # }, # }, # }

Define Tools

tools = [ {'mcpServers': { # You can specify the MCP configuration file "filesystem": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/xxxx/Desktop"] } } } ]

Define Agent

bot = Assistant(llm=llm_cfg, function_list=tools)

Streaming generation

messages = [{'role': 'user', 'content': 'Help me organize my desktop.'}] for responses in bot.run(messages=messages): pass print(responses)

Streaming generation

messages = [{'role': 'user', 'content': 'Develop a dog website and save it on the desktop'}] for responses in bot.run(messages=messages): pass print(responses) ```

Qwen Code

Qwen Code is an open-source AI agent for the terminal, optimized for Qwen models. It helps you understand large codebases, automate tedious work, and ship faster.

For more information, please refer to Qwen Code.

Processing Ultra-Long Texts

Qwen3.6 natively supports context lengths of up to 262,144 tokens. For long-horizon tasks where the total length (including both input and output) exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively., e.g., YaRN.

YaRN is currently supported by several inference frameworks, e.g., transformers, vllm, ktransformers and sglang. In general, there are two approaches to enabling YaRN for supported frameworks:

- Modifying the model configuration file: In the config.json file, change the rope_parameters fields in text_config to: ``json { "mrope_interleaved": true, "mrope_section": [ 11, 11, 10 ], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144, } ``

For vllm, you can use ``shell VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 vllm serve ... --hf-overrides '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --max-model-len 1010000 ``

For sglang and ktransformers, you can use ``shell SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server ... --json-model-override-args '{"text_config": {"rope_parameters": {"mrope_interleaved": true, "mrope_section": [11, 11, 10], "rope_type": "yarn", "rope_theta": 10000000, "partial_rotary_factor": 0.25, "factor": 4.0, "original_max_position_embeddings": 262144}}}' --context-length 1010000 ``

Best Practices

To achieve optimal performance, we recommend the following settings:

1. Sampling Parameters: - We suggest using the following sets of sampling parameters depending on the mode and task type: - Thinking mode for general tasks: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0 - Thinking mode for precise coding tasks (e.g., WebDev): temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0 - Instruct (or non-thinking) mode for general tasks: temperature=0.7, top_p=0.8, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0 - Instruct (or non-thinking) mode for reasoning tasks: temperature=1.0, top_p=1.0, top_k=40, min_p=0.0, presence_penalty=2.0, repetition_penalty=1.0 - For supported frameworks, you can adjust the presence_penalty parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.

2. Adequate Output Length: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 81,920 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.

3. Standardize Output Format: We recommend using prompts to standardize model outputs when benchmarking. - Math Problems: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt. - Multiple-Choice Questions: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the answer field with only the choice letter, e.g., "answer": "C"."

4. Long Video Understanding: To optimize inference efficiency for plain text and images, the size parameter in the released video_preprocessor_config.json is conservatively configured. It is recommended to set the longest_edge parameter in the video_preprocessor_config file to 469,762,048 (corresponding to 224k video tokens) to enable higher frame-rate sampling for hour-scale videos and thereby achieve superior performance. For example, ``json {"longest_edge": 469762048, "shortest_edge": 4096} ``

Alternatively, override the default values via engine startup parameters. For implementation details, refer to: vLLM / SGLang.

Citation

If you find our work helpful, feel free to give us a cite.

``bibtex @misc{qwen36_35b_a3b, title = {{Qwen3.6-35B-A3B}: Agentic Coding Power, Now Open to All}, url = {https://qwen.ai/blog?id=qwen3.6-35b-a3b}, author = {{Qwen Team}}, month = {April}, year = {2026} } ``

Mirrored from the Hugging Face Hub and served from the Conceptio Open Knowledge Archive. Read the original card at https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF.
Legal contextLicense apache-2.0 · EU AI Act — Art. 53 (GPAI model obligations)