[!NOTE]

1. Qwen3.6-35B-A3B-FP8 / Q4_K_M / MTP / GGUF

Qwen3.6-35B-A3B-FP8 transformers converted to .gguf

Quantized to Q4_K_M

MTP recovered and working on latest llama.cpp main repo (natively)

Why FP8 + MTP?

You want near-BF16 fidelity with ~50% memory savings and improved bandwidth efficiency.

This model was exported from FP8 (F8_E4M3) precision provided by Qwen, then quantized to Q4_K_M for GGUF compatibility.

When running with llama.cpp, you get efficient inference via optimized integer kernels and reduced memory footprint.

[!NOTE]

2. This model leverages the FP8 (F8_E4M3) export base to deliver:

- **Near-BF16 fidelity**: Maintains numerical stability for most inference tasks while reducing precision overhead.
- **~50% memory savings**: Q4_K_M uses ~4 bits per weight vs BF16's 16 bits, enabling larger contexts or batch sizes on the same hardware.
- **Improved bandwidth efficiency**: Less data movement translates to higher throughput, especially in memory-bound scenarios.

[!NOTE]

3. Synergy with Multi-Token Prediction (MTP)

When combined with MTP, the reduced memory footprint unlocks additional efficiency gains:
- The lighter weight representation alleviates pressure on the **KV-cache**,
  allowing MTP to speculate more tokens in parallel without OOM risks.
- Efficient dequantization kernels in llama.cpp accelerate inference while maintaining quality.

[!TIP]

To fully benefit from this combination, ensure your inference stack supports:

- A recent build of llama.cpp with CUDA support (`-DGGML_CUDA=ON`)
- MTP-aware decoding (available in latest llama.cpp main branch)
- Sufficient VRAM for your target context length and batch size

[!IMPORTANT] ℹ️ About FP8 + Q4_K_M:
This model was exported from FP8 (F8_E4M3) precision and then quantized to GGUF's Q4_K_M scheme. While the final weights use mixed-precision quantization (mostly ~4-bit), the FP8 export step helps preserve numerical stability for sensitive tensors. The observed speedup comes from reduced memory footprint + efficient GGUF kernels + MTP synergy when running with llama.cpp.

[!CAUTION] Native FP8 acceleration on Ada GPUs is only achievable using the original Transformers checkpoint with a compatible runtime (as documented in Qwen's official FP8 repository).

[!TIP] Enjoy!

[!Note] References:

Base quantization: Qwen3.6 — see Qwen/Qwen3.6-35B-A3B-FP8 for benchmarks.

MTP layers: Taken from havenoammo/Qwen3.6-35B-A3B-MTP-GGUF

convert.py: The script used to graft MTP layers into the GGUFs. Taken from havenoammo/Qwen3.6-35B-A3B-MTP-GGUF

35BA3B-MTP.gguf: The raw Q8_0 MTP layer source file (extracted only) used as input to convert.py.

Thanks to havenoammo and Qwen team

[!TIP] I'm using this settings on RTX4080 16GB:

--ngl 99

--n-cpu-moe 24

-ctk q4_0 -ctv q4_0

--threads 8

-c 178144

--ubatch-size 512 --batch-size 2048

--spec-type draft-mtp --spec-draft-n-max 3 --spec-draft-p-min 0.0

--temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00

--flash-attn on

-np 1

--no-mmap --jinja --no-context-shift

Qwen3.6-27B-FP8

[!Note] This repository contains FP8-quantized model weights and configuration files for the post-trained model in the Hugging Face Transformers format.

These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, KTransformers, etc.

The quantization method is fine-grained fp8 quantization with block size of 128, and its performance metrics are nearly identical to those of the original model.

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

Agentic Coding: the model now handles frontend workflows and repository-level reasoning with greater fluency and precision.
Thinking Preservation: we've introduced a new option to retain reasoning context from historical messages, streamlining iterative development and reducing overhead.

For more details, please refer to our blog post Qwen3.6-27B.

Model Overview

Type: Causal Language Model with Vision Encoder
Training Stage: Pre-training & Post-training
Language Model
- Number of Parameters: 27B
- Hidden Dimension: 5120
- Token Embedding: 248320 (Padded)
- Number of Layers: 64
- Hidden Layout: 16 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN))
- Gated DeltaNet:
  - Number of Linear Attention Heads: 48 for V and 16 for QK
  - Head Dimension: 128
- Gated Attention:
  - Number of Attention Heads: 24 for Q and 4 for KV
  - Head Dimension: 256
  - Rotary Position Embedding Dimension: 64
- Feed Forward Network:
  - Intermediate Dimension: 17408
- LM Output: 248320 (Padded)
- MTP: trained with multi-steps
Context Length: 262,144 natively and extensible up to 1,010,000 tokens.

Benchmark Results

Language

	Qwen3.5-27B	Qwen3.5-397B-A17B	Gemma4-31B	Claude 4.5 Opus	Qwen3.6-35B-A3B	Qwen3.6-27B
Coding Agent
SWE-bench Verified	75.0	76.2	52.0	80.9	73.4	77.2
SWE-bench Pro	51.2	50.9	35.7	57.1	49.5	53.5
SWE-bench Multilingual	69.3	69.3	51.7	77.5	67.2	71.3
Terminal-Bench 2.0	41.6	52.5	42.9	59.3	51.5	59.3
SkillsBench _Avg5	27.2	30.0	23.6	45.3	28.7	48.2
QwenWebBench	1068	1186	1197	1536	1397	1487
NL2Repo	27.3	32.2	15.5	43.2	29.4	36.2
Claw-Eval _Avg	64.3	70.7	48.5	76.6	68.7	72.4
Claw-Eval _Pass^3	46.2	48.1	25.0	59.6	50.0	60.6
QwenClawBench	52.2	51.8	41.7	52.3	52.6	53.4
Knowledge
MMLU-Pro	86.1	87.8	85.2	89.5	85.2	86.2
MMLU-Redux	93.2	94.9	93.7	95.6	93.3	93.5
SuperGPQA	65.6	70.4	65.7	70.6	64.7	66.0
C-Eval	90.5	93.0	82.6	92.2	90.0	91.4
STEM & Reasoning
GPQA Diamond	85.5	88.4	84.3	87.0	86.0	87.8
HLE	24.3	28.7	19.5	30.8	21.4	24.0
LiveCodeBench v6	80.7	83.6	80.0	84.8	80.4	83.9
HMMT Feb 25	92.0	94.8	88.7	92.9	90.7	93.8
HMMT Nov 25	89.8	92.7	87.5	93.3	89.1	90.7
HMMT Feb 26	84.3	87.9	77.2	85.3	83.6	84.3
IMOAnswerBench	79.9	80.9	74.5	84.0	78.9	80.8
AIME26	92.6	93.3	89.2	95.1	92.7	94.1

* SWE-Bench Series: Internal agent scaffold (bash + file-edit tools); temp=1.0, top_p=0.95, 200K context window. We correct some problematic tasks in the public set of SWE-bench Pro and evaluate all baselines on the refined benchmark.
* Terminal-Bench 2.0: Harbor/Terminus-2 harness; 3h timeout, 32 CPU/48 GB RAM; temp=1.0, top_p=0.95, top_k=20, max_tokens=80K, 256K ctx; avg of 5 runs.
* SkillsBench: Evaluated via OpenCode on 78 tasks (self-contained subset, excluding API-dependent tasks); avg of 5 runs.
* NL2Repo: Others are evaluated via Claude Code (temp=1.0, top_p=0.95, max_turns=900).
* QwenClawBench: A real-user-distribution Claw agent benchmark; temp=0.6, 256K ctx.
* QwenWebBench: An internal front-end code generation benchmark; bilingual (EN/CN), 7 categories (Web Design, Web Apps, Games, SVG, Data Visualization, Animation, and 3D); auto-render + multimodal judge (code/visual correctness); BT/Elo rating system.
* AIME 26: We use the full AIME 2026 (I & II), where the scores may differ from Qwen 3.5 notes.

Vision Language

	Qwen3.5-27B	Qwen3.5-397B-A17B	Gemma4-31B	Claude 4.5 Opus	Qwen3.6-35B-A3B	Qwen3.6-27B
STEM & Puzzle
MMMU	82.3	85.0	80.4	80.7	81.7	82.9
MMMU-Pro	75.0	79.0	76.9	70.6	75.3	75.8
MathVista _mini	87.8	--	79.3	--	86.4	87.4
DynaMath	87.7	86.3	79.5	79.7	82.8	85.6
VlmsAreBlind	96.9	--	87.2	--	96.6	97.0
General VQA
RealWorldQA	83.7	83.9	72.3	77.0	85.3	84.1
MMStar	81.0	83.8	77.3	73.2	80.7	81.4
MMBench_EN-DEV-v1.1	92.6	--	90.9	--	92.8	92.3
SimpleVQA	56.0	67.1	52.9	65.7	58.9	56.1
Document Understanding
CharXiv _RQ	79.5	80.8	67.9	68.5	78.0	78.4
CC-OCR	81.0	82.0	75.7	76.9	81.9	81.2
OCRBench	89.4	--	86.1	--	90.0	89.4
Spatial Intelligence
ERQA	60.5	67.5	57.5	46.8	61.8	62.5
CountBench	97.8	97.2	96.1	90.6	96.1	97.8
RefCOCO _avg	90.9	92.3	--	--	92.0	92.5
EmbSpatialBench	84.5	--	--	--	84.3	84.6
RefSpatialBench	67.7	--	4.7	--	64.3	70.0
Video Understanding
VideoMME_{(w sub.)}	87.0	87.5	--	77.7	86.6	87.7
VideoMMMU	82.3	84.7	81.6	84.4	83.7	84.4
MLVU	85.9	86.7	--	81.7	86.2	86.6
MVBench	74.6	77.6	--	67.2	74.6	75.5
Visual Agent
V*	93.7	95.8	--	67.0	90.1	94.7
AndroidWorld	64.2	--	--	--	--	70.3

* Empty cells (--) indicate scores not yet available or not applicable.

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.

[!Important] Inference efficiency and throughput vary significantly across frameworks. We recommend using the latest framework versions to ensure optimal performance and compatibility. For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, KTransformers or vLLM are strongly recommended.

[!Important] The model has a default context length of 262,144 tokens. If you encounter out-of-memory (OOM) errors, consider reducing the context window. However, because Qwen3.6 leverages extended context for complex tasks, we advise maintaining a context length of at least 128K tokens to preserve thinking capabilities.

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:

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.

python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B-FP8 --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.

python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B-FP8 --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:

python -m sglang.launch_server --model-path Qwen/Qwen3.6-27B-FP8 --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:

uv pip install vllm --torch-backend=auto

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.

vllm serve Qwen/Qwen3.6-27B-FP8 --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.

vllm serve Qwen/Qwen3.6-27B-FP8 --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder
Multi-Token Prediction (MTP): The following command is recommended for MTP:

vllm serve Qwen/Qwen3.6-27B-FP8 --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:

vllm serve Qwen/Qwen3.6-27B-FP8 --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:

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:

transformers serve Qwen/Qwen3.6-27B-FP8 --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"

[!Tip] We recommend using the following set of sampling parameters for generation

Thinking mode for general tasks: temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, 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: temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0

Please note that the support for sampling parameters varies according to inference frameworks.

[!Important] Qwen3.6 models operate in thinking mode by default, generating thinking content signified by <think>\n...</think>\n\n before producing the final responses. To disable thinking content and obtain direct response, refer to the examples here.

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/Qwen3.6-27B-FP8",
    messages=messages,
    max_tokens=81920,
    temperature=1.0,
    top_p=0.95,
    presence_penalty=0.0,
    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) $1+\\sqrt{2}$"
            }
        ]
    }
]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.6-27B-FP8",
    messages=messages,
    max_tokens=81920,
    temperature=1.0,
    top_p=0.95,
    presence_penalty=0.0,
    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.
chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.6-27B-FP8",
    messages=messages,
    max_tokens=81920,
    temperature=1.0,
    top_p=0.95,
    presence_penalty=0.0,
    extra_body={
        "top_k": 20,
        "mm_processor_kwargs": {"fps": 2, "do_sample_frames": True},
    }, 
)

print("Chat response:", chat_response)

Instruct (or Non-Thinking) Mode

[!Important] Qwen3.6 does not officially support the soft switch of Qwen3, i.e., /think and /nothink.

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/Qwen3.6-27B-FP8",
    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)

[!Note] If you are using APIs from Alibaba Cloud Model Studio, in addition to changing model, please use "enable_thinking": False instead of "chat_template_kwargs": {"enable_thinking": False}.

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/Qwen3.6-27B-FP8",
    messages=messages,
    max_tokens=32768,
    temperature=0.6,
    top_p=0.95,
    presence_penalty=0.0,
    extra_body={
        "top_k": 20,
        "chat_template_kwargs": {"preserve_thinking": True},
    }, 
)
print("Chat response:", chat_response)

[!Note] If you are using APIs from Alibaba Cloud Model Studio, in addition to changing model, please use "preserve_thinking": True instead of "chat_template_kwargs": {"preserve_thinking": False}.

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': 'qwen3.6-27b',
    '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/Qwen3.6-27B-FP8',
#     '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
1{ 2 "mrope_interleaved": true, 3 "mrope_section": [ 4 11, 5 11, 6 10 7 ], 8 "rope_type": "yarn", 9 "rope_theta": 10000000, 10 "partial_rotary_factor": 0.25, 11 "factor": 4.0, 12 "original_max_position_embeddings": 262144, 13}
Passing command line arguments:

For vllm, you can use

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

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

[!NOTE] All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, potentially impacting performance on shorter texts. We advise modifying the rope_parameters configuration only when processing long contexts is required. It is also recommended to modify the factor as needed. For example, if the typical context length for your application is 524,288 tokens, it would be better to set factor as 2.0.

Best Practices

To achieve optimal performance, we recommend the following settings:

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=0.0, 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:
    temperature=0.7, top_p=0.80, top_k=20, min_p=0.0, presence_penalty=1.5, 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.
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.
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"."
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,

{"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{qwen3.6-27b,
    title  = {{Qwen3.6-27B}: Flagship-Level Coding in a {27B} Dense Model},
    author = {{Qwen Team}},
    month  = {April},
    year   = {2026},
    url    = {https://qwen.ai/blog?id=qwen3.6-27b}
}

eme/Qwen3.6-35B-A3B-FP8-Q4_K_M-MTP-GUFF

1. Qwen3.6-35B-A3B-FP8 / Q4_K_M / MTP / GGUF

2. This model leverages the FP8 (F8_E4M3) export base to deliver:

3. Synergy with Multi-Token Prediction (MTP)

To fully benefit from this combination, ensure your inference stack supports:

Qwen3.6-27B-FP8

Qwen3.6 Highlights

Model Overview

Benchmark Results

Language

Vision Language

Quickstart

Serving Qwen3.6

SGLang

vLLM

KTransformers

Hugging Face Transformers

Using Qwen3.6 via the Chat Completions API

Text-Only Input

Image Input

Video Input

Instruct (or Non-Thinking) Mode

Preserve Thinking

Agentic Usage

Qwen-Agent

Qwen Code

Processing Ultra-Long Texts

Best Practices

Citation

eme/
Qwen3.6-35B-A3B-FP8-Q4_K_M-MTP-GUFF