DavidAU/Qwen3.5-27B-Gemini3-Pro-High-Reasoning-Compact-Thinking

Qwen3.5-27B-Gemini3-Pro-High-Reasoning-Compact-Thinking

EARLY ALPHA.

Fine tune via Unsloth of Qwen 3.5 27B dense model using Gemini distill dataset on local hardware.

This has altered reasoning/thinking block as well as thinking/reasoning block size. (reduced)

Every attempt was made to ensure the training was "mild" and did not negatively affect the model's already incrediblely strong benchmarks.

Vision (images) tested -> working with new training.

BENCHMARKS:

(awaiting finals)

                              arc   arc/e boolq hswag obkqa piqa  wino
THIS MODEL                    0.470,0.550,0.709,...
Qwen3.5-27B-Text-VL qx86-hi   0.443,0.498,0.857,0.701,0.372,0.770,0.752

SAFETY ALIGNMENT:

No attempt was made to adjust/change "censorship" and/or "safety alignment" in the model.

That is coming next. (already built - HERETIC trained models.)

NOTES:

• Gemini Pro Reasoning like "thinking" blocks in most cases.
• Mix of Gemini/Qwen depending on how complex the task(s).
• Suggest min q4ks (non-imatrix) or IQ3S (imatrix).
• Tested with rep pen of 1 (off).
• Context: 256k (default).

Tech notes:

• Due to issues with transformers and/or training there MAYBE issues with the model.
• As this is a NEW MODEL ARCH with it's own quirks regard this tune as "alpha".
• Model may loop in "think" block ; especially Gemini only thinking in some cases ESPECIALLY "mlx" and/or "mxfp4" versions.

[additional updates pending]

IMPORTANT:

• Other versions in testing.
• Information from Qwen's repo below.
• Example generation at the bottom of the page.
• Video portions of the model were NOT TESTED.

---

Qwen3.5-27B

This repository contains 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.

Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency.

Qwen3.5 Highlights

Qwen3.5 features the following enhancement:

• Unified Vision-Language Foundation: Early fusion training on multimodal tokens achieves cross-generational parity with Qwen3 and outperforms Qwen3-VL models across reasoning, coding, agents, and visual understanding benchmarks.

• Efficient Hybrid Architecture: Gated Delta Networks combined with sparse Mixture-of-Experts deliver high-throughput inference with minimal latency and cost overhead.

• Scalable RL Generalization: Reinforcement learning scaled across million-agent environments with progressively complex task distributions for robust real-world adaptability.

• Global Linguistic Coverage: Expanded support to 201 languages and dialects, enabling inclusive, worldwide deployment with nuanced cultural and regional understanding.

• Next-Generation Training Infrastructure: Near-100% multimodal training efficiency compared to text-only training and asynchronous RL frameworks supporting massive-scale agent scaffolds and environment orchestration.

For more details, please refer to our blog post Qwen3.5.

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

	GPT-5-mini 2025-08-07	GPT-OSS-120B	Qwen3-235B-A22B	Qwen3.5-122B-A10B	Qwen3.5-27B	Qwen3.5-35B-A3B
Knowledge
MMLU-Pro	83.7	80.8	84.4	86.7	86.1	85.3
MMLU-Redux	93.7	91.0	93.8	94.0	93.2	93.3
C-Eval	82.2	76.2	92.1	91.9	90.5	90.2
SuperGPQA	58.6	54.6	64.9	67.1	65.6	63.4
Instruction Following
IFEval	93.9	88.9	87.8	93.4	95.0	91.9
IFBench	75.4	69.0	51.7	76.1	76.5	70.2
MultiChallenge	59.0	45.3	50.2	61.5	60.8	60.0
Long Context
AA-LCR	68.0	50.7	60.0	66.9	66.1	58.5
LongBench v2	56.8	48.2	54.8	60.2	60.6	59.0
STEM & Reasoning
HLE w/ CoT	19.4	14.9	18.2	25.3	24.3	22.4
GPQA Diamond	82.8	80.1	81.1	86.6	85.5	84.2
HMMT Feb 25	89.2	90.0	85.1	91.4	92.0	89.0
HMMT Nov 25	84.2	90.0	89.5	90.3	89.8	89.2
Coding
SWE-bench Verified	72.0	62.0	--	72.0	72.4	69.2
Terminal Bench 2	31.9	18.7	--	49.4	41.6	40.5
LiveCodeBench v6	80.5	82.7	75.1	78.9	80.7	74.6
CodeForces	2160	2157	2146	2100	1899	2028
OJBench	40.4	41.5	32.7	39.5	40.1	36.0
FullStackBench en	30.6	58.9	61.1	62.6	60.1	58.1
FullStackBench zh	35.2	60.4	63.1	58.7	57.4	55.0
General Agent
BFCL-V4	55.5	--	54.8	72.2	68.5	67.3
TAU2-Bench	69.8	--	58.5	79.5	79.0	81.2
VITA-Bench	13.9	--	31.6	33.6	41.9	31.9
DeepPlanning	17.9	--	17.1	24.1	22.6	22.8
Search Agent
HLE w/ tool	35.8	19.0	--	47.5	48.5	47.4
Browsecomp	48.1	41.1	--	63.8	61.0	61.0
Browsecomp-zh	49.5	42.9	--	69.9	62.1	69.5
WideSearch	47.2	40.4	--	60.5	61.1	57.1
Seal-0	34.2	45.1	--	44.1	47.2	41.4
Multilingualism
MMMLU	86.2	78.2	83.4	86.7	85.9	85.2
MMLU-ProX	78.5	74.5	77.9	82.2	82.2	81.0
NOVA-63	51.9	51.1	55.4	58.6	58.1	57.1
INCLUDE	81.8	74.0	81.0	82.8	81.6	79.7
Global PIQA	88.5	84.1	85.7	88.4	87.5	86.6
PolyMATH	67.3	54.0	60.1	68.9	71.2	64.4
WMT24++	80.7	74.4	75.8	78.3	77.6	76.3
MAXIFE	85.3	83.7	83.2	87.9	88.0	86.6

* CodeForces: evaluated on our own query set.
* TAU2-Bench: we follow the official setup except for the airline domain, where all models are evaluated by applying the fixes proposed in the Claude Opus 4.5 system card.
* Search Agent: most search agents built on our model adopt a simple context-folding strategy(256k): once the cumulative Tool Response length reaches a preset threshold, earlier Tool Responses are pruned from the history to keep the context within limits.
* WideSearch: we use a 256k context window without any context management.
* MMLU-ProX: we report the averaged accuracy on 29 languages.
* WMT24++: a harder subset of WMT24 after difficulty labeling and rebalancing; we report the averaged scores on 55 languages using XCOMET-XXL.
* MAXIFE: we report the accuracy on English + multilingual original prompts (totally 23 settings).
* Empty cells (--) indicate scores not yet available or not applicable.

Vision Language

	GPT-5-mini 2025-08-07	Claude-Sonnet-4.5	Qwen3-VL-235B-A22B	Qwen3.5-122B-A10B	Qwen3.5-27B	Qwen3.5-35B-A3B
STEM and Puzzle
MMMU	79.0	79.6	80.6	83.9	82.3	81.4
MMMU-Pro	67.3	68.4	69.3	76.9	75.0	75.1
MathVision	71.9	71.1	74.6	86.2	86.0	83.9
Mathvista(mini)	79.1	79.8	85.8	87.4	87.8	86.2
DynaMath	81.4	78.8	82.8	85.9	87.7	85.0
ZEROBench	3	4	4	9	10	8
ZEROBench_sub	27.3	26.3	28.4	36.2	36.2	34.1
VlmsAreBlind	75.8	85.5	79.5	96.7	96.9	97.0
BabyVision	20.9	18.6	22.2	40.2 / 34.5	44.6 / 34.8	38.4 / 29.6
General VQA
RealWorldQA	79.0	70.3	81.3	85.1	83.7	84.1
MMStar	74.1	73.8	78.7	82.9	81.0	81.9
MMBenchEN-DEV-v1.1	86.8	88.3	89.7	92.8	92.6	91.5
SimpleVQA	56.8	57.6	61.3	61.7	56.0	58.3
HallusionBench	63.2	59.9	66.7	67.6	70.0	67.9
Text Recognition and Document Understanding
OmniDocBench1.5	77.0	85.8	84.5	89.8	88.9	89.3
CharXiv(RQ)	68.6	67.2	66.1	77.2	79.5	77.5
MMLongBench-Doc	50.3	--	56.2	59.0	60.2	59.5
CC-OCR	70.8	68.1	81.5	81.8	81.0	80.7
AI2D_TEST	88.2	87.0	89.2	93.3	92.9	92.6
OCRBench	82.1	76.6	87.5	92.1	89.4	91.0
Spatial Intelligence
ERQA	54.0	45.0	52.5	62.0	60.5	64.8
CountBench	91.0	90.0	93.7	97.0	97.8	97.8
RefCOCO(avg)	--	--	91.1	91.3	90.9	89.2
ODInW13	--	--	43.2	44.5	41.1	42.6
EmbSpatialBench	80.7	71.8	84.3	83.9	84.5	83.1
RefSpatialBench	9.0	2.2	69.9	69.3	67.7	63.5
LingoQA	62.4	12.8	66.8	80.8	82.0	79.2
Hypersim	--	--	11.0	12.7	13.0	13.1
SUNRGBD	--	--	34.9	36.2	35.4	33.4
Nuscene	--	--	13.9	15.4	15.2	14.6
Video Understanding
VideoMME(w sub.)	83.5	81.1	83.8	87.3	87.0	86.6
VideoMME(w/o sub.)	78.9	75.3	79.0	83.9	82.8	82.5
VideoMMMU	82.5	77.6	80.0	82.0	82.3	80.4
MLVU	83.3	72.8	83.8	87.3	85.9	85.6
MVBench	--	--	75.2	76.6	74.6	74.8
LVBench	--	--	63.6	74.4	73.6	71.4
MMVU	69.8	70.6	71.1	74.7	73.3	72.3
Visual Agent
ScreenSpot Pro	--	36.2	62.0	70.4	70.3	68.6
OSWorld-Verified	--	61.4	38.1	58.0	56.2	54.5
AndroidWorld	--	--	63.7	66.4	64.2	71.1
Tool Calling
TIR-Bench	24.6	27.6	29.8	53.2 / 42.5	59.8 / 42.3	55.5 / 38.0
V*	71.7	58.6	85.9	93.2 / 90.1	93.7 / 89.0	92.7 / 89.5
Medical VQA
SLAKE	70.5	73.6	54.7	81.6	80.0	78.7
PMC-VQA	36.3	55.9	41.2	63.3	62.4	62.0
MedXpertQA-MM	34.4	54.0	47.6	67.3	62.4	61.4

* MathVision: our model’s score is evaluated using a fixed prompt, e.g., “Please reason step by step, and put your final answer within \boxed{}.” For other models, we report the higher score between runs with and without the \boxed{} formatting.
* BabyVision: scores reported as "with CI / without CI".
* TIR-Bench and V*: scores reported as "with CI / without CI".
* Empty cells (--) indicate scores not yet available or not applicable.

Quickstart

Qwen3.5 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.

For streamlined integration, we recommend using Qwen3.5 via APIs. Below is a guide to use Qwen3.5 via OpenAI-compatible API.

Serving Qwen3.5

Qwen3.5 can be served via APIs with popular inference frameworks. In the following, we show example commands to launch OpenAI-Compatible API servers for Qwen3.5 models.

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.

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 Qwen3.5 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 from the main branch of the open-source repository is required for Qwen3.5, which can be installed using the following command in a fresh environment:

uv pip install 'git+https://github.com/sgl-project/sglang.git#subdirectory=python&egg=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/Qwen3.5-27B --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/Qwen3.5-27B --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/Qwen3.5-27B --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
  

vLLM

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. vLLM from the main branch of the open-source repository is required for Qwen3.5, which can be installed using the following command in a fresh environment:

uv pip install vllm --torch-backend=auto --extra-index-url https://wheels.vllm.ai/nightly

See its documentation for more details.

For detailed Qwen3.5 usage guide, see the vLLM Qwen3.5 recipe.

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/Qwen3.5-27B --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/Qwen3.5-27B --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/Qwen3.5-27B --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/Qwen3.5-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only
  

KTransformers

KTransformers is a flexible framework for experiencing cutting-edge LLM inference optimizations with CPU-GPU heterogeneous computing. For running Qwen3.5 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 Qwen3.5:

pip install "transformers[serving] @ git+https://github.com/huggingface/transformers.git@main"

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 --force-model Qwen/Qwen3.5-27B --port 8000 --continuous-batching

Using Qwen3.5 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.:

pip install -U openai

# Set the following accordingly
export OPENAI_BASE_URL="http://localhost:8000/v1"
export OPENAI_API_KEY="EMPTY"

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=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=0.95, 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.

Text-Only Input

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

messages = [
    {"role": "user", "content": "Type \"I love Qwen3.5\" backwards"},
]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.5-27B",
    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

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/Qwen3.5-27B",
    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

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/Qwen3.5-27B",
    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

Qwen3.5 does not officially support the soft switch of Qwen3, i.e., /think and /nothink.

Qwen3.5 will think by default before response. You can obtain direct response from the model without thinking by configuring the API parameters. For example,

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/RealWorld/RealWorld-04.png"
                }
            },
            {
                "type": "text",
                "text": "Where is this?"
            }
        ]
    }
]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.5-27B",
    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)

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}.

Agentic Usage

Qwen3.5 excels in tool calling capabilities.

Qwen-Agent

We recommend using Qwen-Agent to quickly build Agent applications with Qwen3.5.

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.

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': 'Qwen3.5-27B',
    '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
        },
    },
}

# 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/Qwen3.5-27B',
#     '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}
#         },
#     },
# }

# 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

Qwen3.5 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:

  {
      "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,
  }
  

• 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
  

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:

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. No Thinking Content in History: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. It is implemented in the provided chat template in Jinja2. However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.

5. 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.

@misc{qwen3.5,
    title  = {{Qwen3.5}: Towards Native Multimodal Agents},
    author = {{Qwen Team}},
    month  = {February},
    year   = {2026},
    url    = {https://qwen.ai/blog?id=qwen3.5}
}

---

Example Generation

Q4ks, Lmstudio, Temp .7, rep pen 1

(this will be a low to mid quality generation)

System

User

Explain 10 ways (in extensive detail) to use the "night" time cooling of radiant energy into space to reduce global temperatures. Add charts, icons, emoticons and styling to make it look nice.

Assistant

<think>

Evaluating Radiative Cooling

I'm focused on the core concept: how to leverage nighttime radiative cooling to mitigate global warming. My initial thought is that the Earth loses heat to space at night. I'm now exploring the scientific principles behind this, specifically the "atmospheric window" and how to enhance it. I'm also considering the feasibility of various approaches, from simple material changes to large-scale geoengineering. I'm also thinking about the potential side effects.

Defining the Scope

I'm now zeroing in on the specifics. I'm prioritizing the most promising and practical applications of radiative cooling, while also considering the larger, more theoretical geoengineering approaches. I'm focusing on the scientific mechanisms and how they can be applied on a larger scale to achieve a meaningful reduction in global temperatures. I'm also considering the challenges and limitations of each approach, and how they might be addressed.

Outlining the Strategies

I've solidified the 10 strategies. I'm now working on the detailed explanations for each. I'm making sure to include the scientific basis, how it works, and potential challenges. I'm also integrating the visual elements like charts, icons, and emojis, to make the presentation more engaging. I'm also focusing on the potential impact of each strategy, and how it might contribute to the overall goal of reducing global temperatures.

Detailing the Strategies

I'm now fully immersed in fleshing out the details for each strategy. I'm making sure to explain the scientific principles, the practical applications, and potential challenges for each. I'm also working on the visual elements, like the charts, icons, and emojis, to make the presentation more engaging and easy to understand. I'm focusing on the potential impact of each strategy, and how it might contribute to the overall goal of reducing global temperatures. I'm also considering the scalability and feasibility of each approach.

Refining the Presentation

I'm now polishing the presentation. I'm focusing on the visual elements, ensuring the charts, icons, and emojis are effectively integrated. I'm also revising the language to be more concise and impactful. I'm paying close attention to the formatting to ensure it's visually appealing and easy to read. I'm also considering the overall message and how it's conveyed. I'm aiming for a professional and engaging presentation that clearly communicates the potential of radiative cooling to mitigate global warming.

</think>

🌑 Radiative Sky Cooling: 10 Ways to Lower the Heat

Here is an extensive, detailed breakdown of 10 strategies to utilize Passive Radiative Cooling (PRC). This phenomenon exploits the "Atmospheric Window" (8–13 micrometers), a specific range of infrared wavelengths where the Earth's atmosphere is transparent, allowing heat to escape directly into the freezing vacuum of space.

---

📊 The Science in a Nutshell

Concept	Description
The Vacuum	Space is ~-270°C (-454°F).
The Window	The atmosphere blocks some heat but lets specific IR light pass (8-13 $\mu m$).
The Goal	Emit heat through that window, bypassing the greenhouse effect.

---

1. 🏙️ Super-White Urban Architecture (The "Gorilla" Glue)

Currently, white paint reflects sunlight but absorbs heat at night. We can do better.

The Concept: Standard white paint reflects solar energy (albedo) but is often a poor emitter of thermal infrared energy. By coating roofs and skyscrapers with specialized photonic polymers (like polytetrafluoroethylene or PTFE), we can create surfaces that reflect 99% of sunlight and emit 90%+ of their heat through the atmospheric window.

How it Works: During the day, the building stays cool by reflecting the sun. At night, the material acts as a radiator, dumping heat stored in the concrete and steel directly into space.

The Impact:

• Urban Heat Island Reduction: Can lower building temperatures by 10°C to 20°C below the ambient air temperature.
• Energy Savings: Eliminates the need for AC in mild climates.

> 💡 Pro Tip: This is not just paint; it is a metamaterial engineered to resonate at the exact frequency of the atmospheric window.

---

2. 🚙 Radiative Cooling Highway Coatings

Asphalt acts as a heat sponge. Let's turn it into a heat sink.

The Concept: Asphalt absorbs massive amounts of solar radiation, radiating it back at night, warming the surrounding air. We can replace standard asphalt binders with radiative cooling aggregates.

How it Works: By integrating micro-spheres or specific silicate minerals into the road surface, the road emits thermal radiation at night. This prevents the "heat island" effect where cities stay hot all night.

The Impact:

• Temperature Drop: Can reduce pavement surface temperatures by 30°C during the day and prevent overnight heat retention.
• Air Quality: Cooler roads mean less formation of ground-level ozone (smog).

---

3. 🌾 "Cooling" Agricultural Tents

Protecting crops without water-intensive irrigation.

The Concept: Instead of plastic greenhouses (which trap heat), we build tents using spectrally selective fabrics. These fabrics allow visible light in for photosynthesis but radiate infrared heat out at night.

How it Works: Farmers currently use "shading nets" to reduce heat. Radiative cooling tents go a step further by actively pumping heat out of the crop canopy.

The Impact:

• Water Conservation: Crops transpire less water when temperatures are lower.
• Frost Protection: This method can actually protect crops from frost in specific conditions by maintaining a heat balance, or conversely, prevent heat stress in summer nights.

---

4. 🌊 Oceanic Radiative Desalination

Turning seawater into fresh water using only the night sky.

The Concept: Using floating radiative cooling plates to evaporate seawater at night, then condensing the water vapor.

How it Works:

1. A floating panel cools below the dew point of the water.
2. Water vapor from the ocean rises and condenses on the cold panel.
3. The salt is left behind; the fresh water is collected.

The Impact:

• Zero Energy Cost: No electricity required, just the vacuum of space.
• Scalability: Can be deployed as massive floating arrays in coastal arid regions.

---

5. 🏭 Industrial Radiative Heat Exchangers

Cooling factories using the night sky as a freezer.

The Concept: Factories generate massive heat. Instead of using energy-intensive water cooling towers, we use radiative cooling towers.

How it Works: A network of pipes carries hot factory water to the roof of the plant. The pipes are coated in radiative material. At night, the water flows through these pipes, losing heat to space, and is pumped back down to the factory pre-cooled.

The Impact:

• Carbon Reduction: Drastically reduces the electricity used for industrial cooling.
• Water Efficiency: Eliminates the evaporation loss associated with traditional cooling towers.

---

6. 🛢️ The "Atmospheric Window" Windshield

Cooling cars without the AC compressor running.

The Concept: Car interiors can reach 70°C (158°F). By designing car windshields and roofs with dual-band coatings (visible light transparent, IR transparent, heat emissive).

How it Works: While you are driving during the day, the roof stays cool. At night, the car acts like a radiator. This is particularly useful for parked cars, preventing the "hot car" effect the next morning.

The Impact:

• Fuel Economy: Reduces the load on the car's AC compressor.
• Comfort: Cars are cool to the touch immediately upon entering.

---

7. ☁️ Stratospheric Aerosol Mimicry (The Geoengineering Scale)

Reflecting sunlight, but only at night?

The Concept: This is a more controversial, large-scale geoengineering approach. We could release stratospheric aerosols specifically designed to reflect sunlight during the day, but also designed to be transparent to the night-time cooling window.

How it Works:

• Day: Aerosols reflect sun (Albedo increase).
• Night: Aerosols do not block the escape of heat to space.

The Impact:

• Global Cooling: Could theoretically lower global average temperatures by 0.5°C - 1.0°C.
• Risk: High. Disruption of rainfall patterns is a major concern.

---

8. 🏗️ "Thermal Battery" Concrete

Storing the cold night for the hot day.

The Concept: We can engineer concrete with phase change materials (PCMs) that change state at night.

How it Works:

1. Night: Radiative cooling panels chill the building's concrete core.
2. Storage: The concrete absorbs the cold energy.
3. Day: The concrete releases that stored "coolth" into the building, fighting off the rising sun.

The Impact:

• Time Shift: Allows you to use the "free" night cooling to combat the "expensive" day heat.
• Grid Stability: Reduces peak electricity demand during hot afternoons.

---

9. 🧊 The "Ice Factory" for Developing Nations

Creating ice for food preservation without electricity.

The Concept: In rural areas without electricity, food spoils quickly. We can build radiative cooling ice boxes.

How it Works: A pot of water sits inside a double-walled pot. The outer wall is a super-emitter of infrared radiation. At night, the outer wall radiates heat to space, cooling the water in the inner pot until it freezes into ice.

The Impact:

• Food Security: Allows farmers to store milk, meat, and produce without spoilage.
• Health: Can keep vaccines cool in remote areas.

---

10. 🌬️ The "Sky-Loop" Air Conditioning

A centralized cooling system for whole cities.

The Concept: Imagine a massive central cooling plant located on the outskirts of a city. This plant uses millions of square meters of radiative cooling panels.

How it Works:

1. Cold water is produced at night using the sky.
2. This chilled water is pumped through underground pipes to the city center.
3. Buildings tap into this "sky-chilled" water to run their AC units.

The Impact:

• Decarbonization: Removes the need for electric compressors.
• Scale: One large facility could cool a neighborhood of 10,000 homes.

---

📈 Visualizing the Potential Impact

Cooling Potential by Method

Method	Temp Reduction Potential	Implementation Cost
Roof Coating	-10°C to -20°C	🟢 Low
Road Coating	-15°C (Pavement)	🟡 Medium
Industrial	-5°C (Process Water)	🟡 Medium
Ice Box	-1°C (Freezing Water)	🟢 Low
Geoengineering	-1°C (Global)	🔴 High Risk

---

🧠 Why This Matters

> "We usually try to fight heat with energy. Radiative cooling teaches us to work with physics, not against it."

By utilizing the natural thermodynamics of the night sky, we can transition from energy-intensive cooling (burning gas, using electricity) to energy-generating cooling (using the vacuum of space).

🌍 The Ultimate Goal

To reduce the demand on power grids, we must turn our cities into giant radiators.

🌙 Night: We radiate heat to space. ☀️ Day: We reflect the sun back to space.

The Result: A cooler planet, a cleaner grid, and a more sustainable future. 🚀