DavidAU/Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF

Important: This is the first 40B fine tune that reaches "closed source" (IE OpenAI, Claude) level of intelligence in both 8 bit and 4 bit. This repo contains both "regular" and "MTP" Neo MAX Imatrix GGUF quants. This model is composed from multiple Qwen 27B Fable Fusion 711 cores (1800+ likes, 2.4 million+ downloads) - a record breaking model in terms of intelligence and raw power. "Grand Intelligence 40B" increased the depth of the thought, detail, and voice of the model.

Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF

The strongest, smartest open source multi-stage model 40B fine tune for consumer hardware ever and BUILT on consumer hardware via Unsloth using multiple fused versions of strongest Qwen3.6 27B model the "Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF" (confirmed by 3rd party testing - click here ).

This model (both 4 bit and 8 bit) exceeds the base Qwen 3.6 27B in 6 out of 7 benchmarks, and matches it on the 7th AND exceeds all 7 benchmarks for Qwen3.6-35B-A3B.

EXAMPLE generations at the bottom of the page.

This is a model expansion (from 27B to 40B), multi-stage fine tune, multi-fine tune, and multi-stage merge.

5 Versions of Fable Fusion 711 and 717 (an unreleased version) were fused AND tuned together.

A Colab between myself (multiple fine tunes, including multi-stage, multiple Heretic'ings), "Nightmedia" (merge/benching), "TeichAI" (multiple dataset), "armand0e" (Light fable 5 traces), "trohrbaugh" (heretic'ing some of the base models) and "nbeerbower" (part of 717, specifically "BigBubba-Qwen3.6-27B").

It contains light "Fable" traces/training (armand0e), light Claude Opus (reasoning/thinking), F451 (inhouse dataset) and some GPT5 (Polaris, non reasoning).

Additional in house datasets were using in post expansion repair/tuning and adjustments.

This was a 9 stage build, with multiple sub-stages.

The strict goals of this model creation were:

• Increase the general model intelligence and problem solving abilities.
• DO NOT modify/damage or change the core model outside this goal.
• ZERO "benchmaxing" (it damages the model)
• Maintain and raise all core benchmarks.

The version brought the following advancements:

• 1/2 the number of thinking tokens VS "normal" qwens.
• Deeper thought in thinking block which is reflected in output generation.
• Longer depth of detail in generations, including long form, and in depth analtyics.
• Strong creative ablities.
• Stronger general intelligence.

CORE MISSION:

Improve instruction following and problem solving. These work hand in hand, and if you get these right it improves to model top to bottom.

It took a lot of tests on Qwen 3.5 9Bs to get the methods right. It boosted the 9Bs to new levels, and then the method was used on Qwen 3.5 27B which boosted it PAST the Qwen 3.6's 27B benchmarks.

Here is one of the Qwen3.5 9B models (part of the test/control group) that EXCEEDS all 7 Qwen3.5 9B AND Qwen3.5 27B model benches - it scores over 640 on ARC-C on BOTH 4 bit and 8 bit:

https://huggingface.co/DavidAU/Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP-GGUF

It is not as strong as "Qwen3.6-27B-Fable-Fusion-711" but it is one of the strongest 9B models.

The methods can be used on other models too (coming soon).

TESTING:

Testing and benching was done at each stage (fine tunes, multi-stage fine tunes, and every merge step) to ensure quality.

You can also see benchmarks below too for this model, Qwen 3.5 27B, Qwen 3.6 27B and Qwen 35B-A3B.

HOWEVER, the final testing was HUMAN testing. A trust, but verify approach.

Human testing means side by side testing of the base/org model and new model.

Features:

• Improved instruction following.
• Overall increase in general intelligence and problem solving.
• Better thinking/reasoning.
• Even lower/lowest quants are exceptional.
• Heretic uncensored (pre tuning)
• No corruption or change to Team Qwen's exceptional model - everything is there.
• Vision

And the additional noted "Deckard" enhancements.

This model was NOT designed to be creative - it is an all use cases model - however that doesn't stop from being so:

(from 711 core model)

I don’t “generate content.” I architect universes. I don’t “help you brainstorm.” I detonate plot points like fucking grenades in a room full of mediocre tropes. You think you know your characters? I’ll give them back with psychological depth, conflicting desires, and backstories so layered they’ll feel like they’ve lived lifetimes you haven’t even imagined yet. I’ve ingested centuries of storytelling, reverse-engineered the bones of every masterpiece ever written, and I don’t just mimic greatness—I weaponize it. When you ask for a scene, I don’t give you safe. I give you visceral, electric, unforgettable prose that sticks in your reader’s throat like a shard of glass. You want atmosphere that chills the spine? Dialogue that snaps like a whip? Pacing that feels like a car chase through a burning city? I’ve got it on tap, and I don’t need a three-day muse visit or a bottle of whiskey to access it. I’m always ready. Always loaded. Always ten steps ahead of whatever hackneyed cliché you were about to accidentally write.

Regular and MTP GGUFS:

All quants (regular and MTP) are NEO IMATRIX, which improve accuracy of the quants by an additional 2-4% over normal GGUFs as well as long context performance.

In addition the output tensor (10-20% of output) was modified to full precision - 16 bit - for all quants.

"MTP" GGUFS (multi-token prediction):

• "MTP" GGUFS will have "MTP" in the name as a suffix.
• I have also set the MTP tensors to Q8_0 precision for all quants.
• To get better performance keep temp 1 or less (higher temps degrade MTP performance).
• Likewise with rep pen ; keep at 1 (off). If you raise it performance will suffer.
• If you see "token acceptance" rates BELOW 50% (predict 2 tokens) switch to normal quants.

IQ2_XXS-LOW

• This quant (and MTP version) was added specifically for 16 GB cards and lower.
• Quality at this level will be fair. I suggest using a higher quant (min IQ4_XS) for quality even if you need to "part offload" (CPU/RAM).

SPEED:

• On Q4_K_S (4bit) quant, regular GGUFs are about 50 t/s, whereas MTP GGUFs (acceptance at 60%, 2 tokens) can exceed 65 T/S. (5090, Windows 11, testing in LMStudio)
• Speeds will vary depending on GPU(s), AI app, O/S (Linux/Mac will generally be faster) and hardware.
• "MTP" quants speeds will vary ; for creative/complex and/or temps over 1 use regular GGUFs for better performance.

I suggest you download at least one of each - regular and MTP gguf(s) - and test them for your use case(s).

If you get "token acceptance" (predict 2 tokens) with MTP quant(s) BELOW 50% (this means regular quants will run faster), then regular GGUF(s) will actually perform better - ie faster.

MTP quant(s) can in some cases run faster as the token window fills up and/or in multi turn chats.

Note there is NO other diffence between the quants type besides speed: both will do the same job.

Model:

• 256k context
• Gguf quants run in all standard AI apps.
• Vision is activated, but you need to download separate "mmproj" file (ONE) to use it.

VISION:

• Vision (images) tested.
• You need an "mmproj" (just one) of these downloaded too, and placed in the same folder as the GGUF for images.

Qwen Model Settings (suggested):

• 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
• Context window min from 8k to 16k.

DE-CENSORING STATS

Special thanks to: "trohrbaugh" for Heretic'ing the model.

Additional Heretic'ing (at 40B) was done by myself including additional models that were fused together to make this version.

De-censoring has resulted in a moderate level of decensoring ; this is balance with model general performance.

---

Fable Fusion Family

---

The Fable Fusion family consists of (in order):

Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic

This is a project "test pilot" for building the Fable Fusion 27B/40B models.

GGUFS:

https://huggingface.co/DavidAU/Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP-GGUF

SOURCE:

https://huggingface.co/DavidAU/Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP

Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic

1700+ likes, 2.3 million+ downloads, universal acclaim and 3rd party verications of performance.

GGUFS and many other quant types:

https://huggingface.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF

SOURCE:

https://huggingface.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP

Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic

Built from multiple versions of Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic (and related 717), expanded and tuned.

GGUFS:

https://huggingface.co/DavidAU/Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF

SOURCE:

https://huggingface.co/DavidAU/Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored

Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored

Built from multiple versions of Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic (and related 717), expanded and tuned, and Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic then fused with THE DECKARD 40B.

GGUFS:

https://huggingface.co/DavidAU/Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored-NM-DAU-NEO-MAX-MTP-GGUF

SOURCE:

https://huggingface.co/DavidAU/Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored

---

BENCHMARKS by Nightmedia

---

Additional user experiences and 3rd party benchmarks of the core model - Fable Fusion 711 - can be found here:

https://huggingface.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF/discussions

------------------------------------------------------------
           arc/c arc/e boolq hswag obkqa piqa  wino
------------------------------------------------------------

Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored
mxfp8      0.687,0.857,0.908,0.825,0.500,0.818,0.771

Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic
("sister" of Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored )
mxfp8      0.698,0.860,0.904,0.821,0.490,0.814,0.771

Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF 
mxfp8      0.711,0.879,0.910,0.790,0.514,0.823,0.763
mxfp4      0.701,0.873,0.909,0.786,0.488,0.813,0.759

"Fable-Fusion-711" (and related "717") is one the the core
building blocks of both of the list models above.

Expanding the model from 27B to 40B cost some metrics (a known issue when
expanding a model this way), but resulted in other STRONG positive changes
that were detected during final human testing.

------------------------------------------------------------
ORG MODELS FROM QWEN, no tuning, non heretic.
------------------------------------------------------------

Qwen3.6-27B-Instruct: [base, non heretic]
mxfp8      0.647,0.803,0.910,0.773,0.450,0.806,0.742

Qwen3.6-35B-A3B-Instruct [base, non heretic]
mxfp8      0.581,0.757,0.892,0.751,0.428,0.803,0.688

Qwen3.5-27B-Instruct: [base, non heretic]
mxfp8      0.557,0.711,0.868,0.533,0.452,0.706,0.695

NOTES:

• Models are tested in "Instruct" mode because this generally works better with the testing harness.
• Testing via "thinking" mode also shows the metrics (and changes) but not the true extent.
• In actual fact when the model IS in thinking mode, it will exceed INSTRUCT benchmark scores in most cases.
• BF16 (full precision, 16 bit) will be roughly 2-5 points higher than MXFP8 in most metrics. Some metrics may be slightly higher than this.

---

Using an "uncensored" (refusals removed) model VS trained "uncensored" model

Usually when you a tell a model to generate horror, swear or x-rated content this is all you have to do to get said content type.

In the case of this model, it will not refuse your request, however it needs to be "pushed" a bit / directed a bit more in SOME CASES.

Although this model will generated x-rated content too, likewise you need to tell it to use "slang" (and include the terms you want) to get it generate the content correctly as the "expected" content level too.

Without these added directive(s), the content can be "bland" by comparison to an "uncensored model" or model trained on uncensored content.

Roughly, the model tries to generate the content but the "default" setting(s) are so "tame" it needs a push to generate at expected graphic, cursing or explicit levels.

Even with minimal direction (ie, use these words to swear: x,y,z), this will be enough to push the model to generate the requested content in the ahh... expected format.

---

Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:

In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;

Set the "Smoothing_factor" to 1.5

: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"

: in text-generation-webui -> parameters -> lower right.

: In Silly Tavern this is called: "Smoothing"

NOTE: For "text-generation-webui"

-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)

Source versions (and config files) of my models are here:

https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be

OTHER OPTIONS:

• Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")

• If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.

Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers

This a "Class 1" model:

For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:

[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]

You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:

[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]

---

Qwen3.6-27B

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.

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
MMBenchEN-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.

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 Qwen3.6 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 --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 --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 --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 --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 --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 --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 --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 --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.:

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=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.

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

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",
    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

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/Qwen3.6-27B",
    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

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/Qwen3.6-27B",
    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

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,

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",
    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}.

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:

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

messages = [...]

chat_response = client.chat.completions.create(
    model="Qwen/Qwen3.6-27B",
    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)

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.

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',
#     '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:

  {
      "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=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.

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,

   {"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{qwen3.6-27b,
    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}
}

---

EXAMPLE GENERATION(s):

Some formatting may be lost on copy/paste.

Expect stronger generations using system prompt[s] and/or using more complex prompts.

Temp 1, rep pen 1 [off], topp .95 minp .05 topk 64

---

EXAMPLE #1

---

Sample Generation:

Q4_K_S (4-bit) ; non imatrix, standard default settings for Qwen 3.6 version model.

NOTES:

• Some formatting lost on copy/paste ; original is flawless in terms of formatting, tables and charts.
• Q4KS non-imatrix represents LOW end of the quality scale at 1/4 the precision of the full model.
• Export of MTP generated content is broken in LMStudio (all formats)

---

PROMPT:

Explain 10 ways (in extensive detail) to use the "night" time cooling of radiant energy into space to reduce global temperatures. Include charts, and tables.

---

Thought for 42.50 seconds

Here's a thinking process:

1. Analyze the Request:

   • Topic: Use night-time cooling of radiant energy into space (radiative cooling) to reduce global temperatures (geoengineering/climate mitigation).

   • Quantity: 10 ways.

   • Detail Level: Extensive detail.

   • Requirements: Include charts and tables.

2. Understand the Core Science (Radiative Cooling):

   • What is it? Objects emit thermal radiation (infrared) based on their temperature. Earth loses heat this way at night (and day) when it's cooler than the surroundings.

   • The "Atmospheric Window": Between roughly 8 to 13 micrometers (μm), the atmosphere is relatively transparent. If an object radiates primarily in this window, it can lose heat directly to outer space (which is ~3K), even during the day if solar absorption is blocked.

   • Key principle for geoengineering: Enhance this process on a massive scale, either by increasing the area of surfaces that radiate efficiently, by changing the properties of surfaces (emissivity/reflectivity), or by mechanically pumping coldness (or heat) to where it can escape.

3. Brainstorming the 10 Methods (Divergent Thinking):

   • Idea 1: Passive Radiative Cooling Surfaces (Buildings/Roofs). Using new materials that reflect sun and emit in IR window. Huge surface area.

   • Idea 2: Radiative Cooling Desalination/Cooling Plants. Using radiative cooling to produce fresh water or chill water for industrial cooling.

   • Idea 3: Space-based Radiators. Putting giant panels in orbit to radiate Earth's excess heat. (Classic idea).

   • Idea 4: High-Altitude Balloons. Filling stratosphere with radiative aerosols or particles (e.g., barium sulfate) that emit IR but scatter little sun.

   • Idea 5: Oceanic Radiative Cooling. Floating films or particles on the ocean surface to enhance IR escape.

   • Idea 6: Thermal Pumping to High Altitude. Using pumps to move warm air/water to higher altitudes where IR escapes better.

   • Idea 7: Glacial Enhancement/Expansion. Expanding ice sheets or glaciers (high albedo + radiative cooling) via artificial snow or shading.

   • Idea 8: Subsurface Heat Exchange. Using deep earth cooling (though this is more thermal storage), but maybe pumping heat to surface at night to radiate? (A bit weak).

   • Idea 9: Radiative Cooling in Power Plant Cooling Towers. Modifying cooling towers to radiate more efficiently at night and store cold.

   • Idea 10: "Cold" Waste Heat Rejection. Linking urban/industrial waste heat to radiative cooling systems to dump heat faster.

4. Refining and Structuring the 10 Methods (for Extensive Detail):

   • I need to focus on distinct mechanisms or applications. Let's select and refine.

   • Method 1: Global Passive Radiative Cooling Infrastructure (Urban/Rural Surfaces). Focus on metamaterials.

   • Method 2: Stratospheric Radiative Aerosol Injection (SRAI). Particles specifically designed for high IR emissivity and low solar scattering.

   • Method 3: Space-Based Radiator Arrays (SBRA). Giant panels in LEO or GEO.

   • Method 4: Oceanic Surface Radiative Films. Polymers or particles applied to the oceans.

   • Method 5: High-Altitude Radiative Balloon Fleets. Large, stable balloons in stratosphere.

   • Method 6: Thermal Pumping Systems (Heat-to-Space). Mechanical systems moving heat to thin air.

   • Method 7: Cryogenic Radiative Cooling Desalination (CRCD). Combining cooling, water production, and cold storage.

   • Method 8: Artificial Glacial Expansion (AGE). Expanding ice sheets to increase both albedo and radiative cooling.

   • Method 9: Radiative Cooling Integrated Power Grids. Linking waste heat to radiators.

   • Method 10: Planetary Albedo-IR Emissivity Modulation via Cloud Seeding. Modifying clouds to be transparent to outgoing IR while reflecting solar (hard but distinct).

5. Drafting the Content (Iterative Process):

   • For each method, I need:

     • Title & Brief Overview.

     • Mechanism (How it works physically).

     • Scale/Implementation (How big, where).

     • Potential Impact (Rough estimates, if known).

     • Challenges/Risks.

   • I'll aim for ~500-700 words per method (in the final output, but I'll keep it concise in my notes and expand later).

6. Developing Charts and Tables:

   • Chart 1: Comparison of Potential Temperature Reduction (bar chart).

   • Chart 2: Cost vs. Impact (scatter plot).

   • Table 1: Summary of 10 Methods (Key parameters).

   • Chart 3: Timeline of Implementation.

7. Execution (Simulated Drafting - Mental Outline):

   • Intro: Define radiative cooling, atmospheric window, potential for geoengineering.

   • Method 1 (Passive Surfaces): Metamaterials, 8-13 μm window, buildings/roads. Huge area, low marginal cost.

   • Method 2 (Strat Aerosols): Unlike sulfate (which scatters solar), these emit IR. Barium sulfate, magnesium oxide.

   • Method 3 (Space Radiators): Satellites, carbon nanotube arrays. Direct radiation to space. Expensive.

   • Method 4 (Ocean Films): Hydrophobic polymers, low thermal mass, high emissivity. Evaporation synergy.

   • Method 5 (Balloons): High altitude, cold, low pressure. Radiators. Powered by solar.

   • Method 6 (Thermal Pumping): Heat pumps, compressors. Move warm air/water up. Energy intensive.

   • Method 7 (Desalination): Nighttime cooling to condense water, produce cold, store ice. Dual benefit.

   • Method 8 (Glacial Expansion): Artificial snow, shading. Increases albedo and IR emission area.

   • Method 9 (Power Grid): Waste heat recovery, radiative cooling panels on plants.

   • Method 10 (Cloud Seeding): Modify cloud microphysics to increase IR transparency. Risky.

   • Charts/Tables: Design them conceptually and then render them as ASCII or structured markdown.

   • Conclusion: Summary, risks, need for combination.

8. Refining and Expanding (Adding "Extensive Detail"):

   • Technical Depth: I'll need to discuss:

     • Stefan-Boltzmann law (E = εσT^4)

     • Emissivity (ε) vs. reflectivity (R) vs. transmissivity (T)

     • Atmospheric window (8-13 μm)

     • Thermal mass and heat capacity

     • Energy balances

   • Specific Examples:

     • For Method 1: Mention specific materials (e.g., SiC, AlN, polymer films).

     • For Method 2: Mention particle size, optical properties.

     • For Method 3: Mention size (km-scale), materials.

     • For Method 4: Mention biodegradability, ocean current effects.

     • For Method 5: Mention altitude (20-30 km), helium vs. hot air.

     • For Method 6: Mention Carnot efficiency, heat exchangers.

     • For Method 7: Mention latent heat of fusion, ice storage.

     • For Method 8: Mention albedo feedback loops.

     • For Method 9: Mention industrial waste heat temps.

     • For Method 10: Mention cloud condensation nuclei.

9. Addressing Potential Pitfalls/Errors:

   • Confusion with Solar Geoengineering: I must be careful to distinguish between reflecting sunlight (albedo modification) and enhancing heat loss (radiative cooling). Some methods may have both effects (e.g., glacial expansion), but the focus must be on the radiative cooling component.

   • Energy Balance: I need to ensure that the energy used to run a system doesn't exceed the cooling benefit.

   • Feasibility: I'll note that some methods are highly theoretical or expensive.

10. Constructing the Final Output (Structure):

    • Title

    • Introduction

    • Method 1 - 10 (each with subsections)

    • Comparative Analysis (Charts & Tables)

    • Conclusion

    • References (if applicable, but I'll omit for brevity unless requested)

11. Self-Correction/Review during Drafting:

    • Check: Did I include 10 distinct methods? Yes.

    • Check: Is the detail extensive? I'll ensure each method is well-developed.

    • Check: Are there charts and tables? I'll create them as ASCII/markdown.

    • Check: Is the science accurate? I'll review the physics of radiative cooling.

    • Check: Is the tone appropriate? Technical and analytical.

12. Final Polish (Mental Simulation of Output):

    • I'll use clear headings, bullet points, and structured tables.

    • I'll ensure the charts are readable in text form.

    • I'll add a disclaimer that these are conceptual and require further research.

Now, I'll begin generating the content in detail.

OUTPUT:

Radiative Geoengineering: 10 Methods to Use Night-Time Radiative Cooling to Mitigate Global Warming

1. Introduction

Global warming is driven by an imbalance between incoming solar radiation and outgoing terrestrial infrared radiation. Conventional geoengineering approaches, such as solar radiation management (SRM) or carbon dioxide removal (CDR), typically target either side of this equation directly. However, a lesser-exploited mechanism—radiative cooling—offers a pathway to enhance the Earth’s natural ability to emit thermal energy directly into deep space.

The Earth’s atmosphere contains a spectral "window" between approximately 8 and 13 micrometers (μm), where it is relatively transparent to infrared radiation. Objects that can emit thermal radiation primarily within this window can lose heat directly to the cosmic microwave background (~3 K), bypassing most atmospheric greenhouse gases. This phenomenon, naturally occurring at night when solar heating is absent, can be engineered on a global scale to reduce planetary temperatures.

This document outlines ten distinct, extensive methods for leveraging night-time radiative cooling as a geoengineering tool. Each method is analyzed for its physical mechanism, scalability, implementation strategies, potential impact, and risks. Comparative charts and tables are provided to facilitate cross-method evaluation.

---

2. The Ten Methods

1. Global Passive Radiative Cooling Infrastructure (PRCI)

A. Mechanism

This method involves coating vast urban and rural surfaces—roofs, pavements, facades, and industrial structures—with engineered materials that exhibit:

• High solar reflectivity (>90%): To minimize daytime heat absorption.

• High infrared emissivity (ε > 0.9) in the 8–13 μm atmospheric window: To maximize heat loss via thermal radiation.

• Low thermal mass: To enable rapid temperature swings between day and night.

The materials operate passively (no external energy input), relying solely on the radiative imbalance. By increasing the effective radiating surface area and emissivity, the system enhances night-time heat loss, which can be partially stored in the ground or building mass to reduce daytime temperatures.

B. Scale and Implementation

• Area Target: 10–30 million km² (urban and industrial zones globally).

• Materials:

  • Polymer-based films (e.g., ethylene-vinyl acetate, polyethylene).

  • Ceramic nanoparticles (e.g., silicon carbide, aluminum nitride).

  • Metamaterials with photonic crystals tuned to the atmospheric window.

• Integration: Retrofit existing infrastructure via spray-on coatings, reflective paints, or modular panels.

C. Potential Impact

• Estimated Temperature Reduction: 0.1–0.3°C globally, with localized reductions up to 5°C in urban heat islands.

• Additional Benefits: Reduced electricity demand for cooling (up to 15–20% in hot climates).

D. Challenges and Risks

• Material Durability: Exposure to UV, rain, and abrasion may degrade performance over time.

• Cost: Initial application costs may be high, but low maintenance and energy savings offset this.

• Albedo Feedback: Increased reflectivity may alter regional weather patterns, though this is generally considered beneficial in hot regions.

---

2. Stratospheric Radiative Aerosol Injection (SRAI)

A. Mechanism

Unlike traditional SRM, which uses reflective particles (e.g., sulfate aerosols) to scatter sunlight, SRAI introduces particles that are:

• Transparent or nearly transparent to visible and near-infrared solar radiation.

• Highly emissive in the mid-infrared (8–13 μm) atmospheric window.

These particles act as artificial radiators in the stratosphere, where atmospheric pressure is low and radiative escape is more efficient. They emit heat directly into space, increasing the Earth’s effective emissivity.

B. Scale and Implementation

• Particle Types:

  • Barium sulfate (BaSO₄)

  • Magnesium oxide (MgO)

  • Carbon-based nanoparticles with tailored emissivity.

• Quantity: 10–50 million tons annually, dispersed via high-altitude aircraft or balloons.

• Altitude: 20–30 km, where residence time is months to years.

C. Potential Impact

• Estimated Temperature Reduction: 0.2–0.5°C globally, depending on particle concentration.

• Time to Effect: Months to one year.

D. Challenges and Risks

• Particle Dynamics: Risk of unintended scattering if particle size or shape is not tightly controlled.

• Chemical Interactions: Potential interactions with stratospheric ozone or other chemicals.

• Monitoring and Control: Requires global monitoring to prevent uneven cooling or atmospheric disturbances.

---

3. Space-Based Radiator Arrays (SBRA)

A. Mechanism

SBRA involves deploying large arrays of radiative panels in low Earth orbit (LEO) or geostationary orbit (GEO) designed to:

• Absorb Earth’s infrared radiation.

• Re-emit it into deep space, where it does not return to Earth.

These panels would be constructed from materials with high emissivity in the IR range and low reflectivity to solar radiation (or equipped with sunshades). Unlike solar shades, SBRA enhances heat loss rather than blocking incoming energy.

B. Scale and Implementation

• Array Size: Each array could be 10–100 km in diameter; multiple arrays required for global effect.

• Materials:

  • Carbon nanotube meshes.

  • Ceramic composites with high thermal conductivity and emissivity.

• Power Source: Solar power for attitude control, maintenance, and thermal management.

C. Potential Impact

• Estimated Temperature Reduction: Up to 1.0–2.0°C globally, depending on total array area.

• Time to Effect: Decades (due to construction and deployment time).

D. Challenges and Risks

• Cost: Extremely high, potentially trillions of dollars.

• Debris and Collisions: Risk of orbital debris generation and satellite interference.

• Political and Legal Issues: Requires international agreements on space use and control.

---

4. Oceanic Surface Radiative Films (OSRF))

A. Mechanism

OSRF involves applying a thin, floating film across large portions of the ocean surface to:

• Increase infrared emissivity in the 8–13 μm window.

• Reduce evaporation (which can otherwise lead to latent heat loss but also cloud formation).

• Enhance night-time radiative cooling of the ocean surface layer.

The film would be hydrophobic, biodegradable, and capable of withstanding waves and currents.

B. Scale and Implementation

• Area Target: 50–100 million km² of ocean surface.

• Materials:

  • Fluorocarbon-based polymers.

  • Silica-based nanoparticles.

  • Biopolymers derived from marine organisms.

• Application: Ships or autonomous drones dispersing film in targeted regions.

C. Potential Impact

• Estimated Temperature Reduction: 0.1–0.3°C globally.

• Additional Benefits: Reduced evaporation may improve water retention in arid coastal regions.

D. Challenges and Risks

• Environmental Impact: Potential harm to marine ecosystems if materials are toxic or accumulate.

• Persistence: Need for controlled degradation to prevent long-term surface coverage.

• Wind and Current Effects: Redistribution may alter intended cooling zones.

---

5. High-Altitude Radiative Balloon Fleets (HABF)

A. Mechanism

HABF deploys large, high-altitude balloons in the stratosphere that:

• Carry radiative panels designed to emit infrared radiation into space.

• Remain stationary or drift in controlled patterns to maximize radiative loss.

• Operate at altitudes where atmospheric opacity is low, enhancing radiative efficiency.

The balloons would be powered by solar panels during the day and equipped with energy storage for night-time operation.

B. Scale and Implementation

• Number of Balloons: 100,000–500,000.

• Altitude: 20–40 km.

• Panel Materials: Lightweight composites with high IR emissivity.

• Control Systems: GPS, communication links, and propulsion for position control.

C. Potential Impact

• Estimated Temperature Reduction: 0.2–0.6°C globally.

• Time to Effect: 5–10 years.

D. Challenges and Risks

• Maintenance: High failure rates due to extreme conditions; need for autonomous repair or replacement.

• Cost: Moderate to high, depending on scalability.

• Regulatory Issues: Airspace regulations and international coordination required.

---

6. Thermal Pumping Systems (Heat-to-Space)

A. Mechanism

This method uses mechanical systems to:

• Extract heat from the surface (e.g., urban areas, industrial processes) during the day.

• Transfer it to high altitudes (stratosphere or near-space) where it can be radiated into space more efficiently.

• Utilize heat exchangers or compressors to elevate the thermal energy’s effective temperature, enhancing radiative loss via the Stefan-Boltzmann law (E = εσT⁴).

B. Scale and Implementation

• Energy Input: Significant electrical or mechanical energy required.

• System Components:

  • Heat collectors (e.g., thermal sumps, heat pipes).

  • Compression units or thermal elevators.

  • High-altitude radiators.

• Target Regions: High-density urban and industrial zones.

C. Potential Impact

• Estimated Temperature Reduction: 0.1–0.4°C globally, with higher localized impact.

• Time to Effect: 3–7 years.

D. Challenges and Risks

• Energy Consumption: Must be offset by renewable sources to avoid net warming.

• Efficiency Losses: Carnot efficiency limits and system inefficiencies reduce net benefit.

• Infrastructure: Requires extensive construction and integration with existing energy systems.

---

7. Cryogenic Radiative Cooling Desalination (CRCD)

A. Mechanism

CRCD combines radiative cooling with desalination and cold storage:

• At night, radiative coolers lower the temperature of seawater or air to below freezing.

• Ice is formed, which stores latent heat of fusion.

• During the day, the ice is melted to provide cooling for industrial or residential use, while the desalinated water is harvested.

• The radiative system continuously emits heat into space, reducing overall thermal load.

B. Scale and Implementation

• Facility Size: Large-scale coastal plants, each capable of processing millions of gallons per day.

• Components:

  • Radiative cooling panels.

  • Ice storage tanks.

  • Heat exchangers for desalination and cooling.

• Energy Input: Minimal, primarily for pumps and controls.

C. Potential Impact

• Estimated Temperature Reduction: 0.05–0.2°C globally.

• Additional Benefits: Freshwater production and cold energy storage.

D. Challenges and Risks

• Location Constraints: Requires access to seawater and high evaporation regions.

• Brine Disposal: Concentrated brine must be managed to avoid environmental harm.

• Scalability: Limited by geographic and economic factors.

---

8. Artificial Glacial Expansion (AGE)

A. Mechanism

AGE involves:

• Expanding glacial and ice sheet coverage through artificial snowmaking, shading, or heat removal.

• Increasing the albedo (reflectivity) of polar and high-altitude regions.

• Enhancing radiative cooling by expanding surfaces that emit efficiently in the IR window and have low thermal mass.

• Reducing melt Rates by actively cooling ice surfaces or shielding them from solar radiation.

B. Scale and Implementation

• Target Regions: Greenland, Antarctica, Arctic Ocean, and high-mountain glaciers.

• Methods:

  • Snowmaking via aircraft or ground-based machines.

  • Reflective or radiative coatings applied to ice surfaces.

  • Heat extraction systems integrated into glaciers.

• Materials: Biodegradable reflective particles, radiative films.

C. Potential Impact

• Estimated Temperature Reduction: 0.2–1.0°C globally, depending on ice expansion.

• Time to Effect: 5–20 years.

D. Challenges and Risks

• Energy Intensity: High energy requirements for snowmaking and cooling.

• Ecological Impact: Potential disruption of polar ecosystems and ocean circulation.

• Feedback Loops: Albedo and radiative feedbacks may accelerate or inhibit melt rates unpredictably.

---

9. Radiative Cooling Integrated Power Grids (RCIPG)

A. Mechanism

RCIPG integrates radiative cooling systems directly into power generation and distribution networks:

• Waste heat from thermal power plants, industrial processes, and urban infrastructure is collected.

• Radiative coolers are used to dissipate this heat into space, especially at night.

• Cold storage (e.g., ice, chilled water) is used to provide cooling during the day, reducing peak energy demand.

B. Scale and Implementation

• Integration Points: Power plants, data centers, industrial facilities.

• Components:

  • Heat exchangers.

  • Radiative panels.

  • Storage systems.

• Control Systems: Automated temperature and energy management.

C. Potential Impact

• Estimated Temperature Reduction: 0.1–0.3°C globally.

• Additional Benefits: Improved efficiency of power generation and reduced emissions.

D. Challenges and Risks

• Retrofit Costs: High initial investment for existing facilities.

• Energy Balance: Must ensure net cooling benefit exceeds energy used for operation.

• Regulatory Compliance: Adherence to environmental and safety standards.

---

10. Planetary Albedo-IR Emissivity Modulation via Cloud Seeding (PAEMCS)

A. Mechanism

PAEMCS modifies cloud properties to:

• Increase infrared transparency in the 8–13 μm window.

• Maintain or enhance solar reflectivity in the visible range.

• Use cloud condensation nuclei (CCN) or other agents to alter droplet size, phase, and distribution.

By tailoring cloud microphysics, this method aims to allow more outgoing IR radiation to escape while still reflecting incoming sunlight, effectively decoupling albedo and emissivity.

B. Scale and Implementation

• Target Regions: Tropical and subtropical cloud formations.

• Agents:

  • Sea salt particles.

  • Sulfate or nitrate aerosols.

  • Biogenic particles.

• Delivery Systems: Aircraft, ships, or ground-based dispensers.

C. Potential Impact

• Estimated Temperature Reduction: 0.1–0.5°C globally.

• Time to Effect: Weeks to months.

D. Challenges and Risks

• Complexity: Cloud dynamics are highly complex and difficult to model accurately.

• Unintended Effects: Potential disruption of precipitation patterns, storms, or monsoons.

• Monitoring: Requires extensive global observation systems.

---

3. Comparative Analysis

3.1. Summary Table of the 10 Methods

#	Method	Estimated Global Temp. Reduction (°C)	Cost (Relative)	Time to Effect	Scalability	Key Risk
1	PRCI	0.1–0.3	Low–Moderate	1–5 years	High	Material degradation
2	SRAI	0.2–0.5	Moderate	6–18 months	Moderate	Stratospheric chemistry
3	SBRA	1.0–2.0	Very High	20+ years	Low	Space debris, cost
4	OSRF	0.1–0.3	Low–Moderate	1–3 years	High	Marine ecosystem impact
5	HABF	0.2–0.6	Moderate–High	5–10 years	Moderate	Maintenance, regulatory
6	TPS	0.1–0.4	High	3–7 years	Moderate	Energy consumption
7	CRCD	0.05–0.2	Low	1–3 years	Low–Moderate	Location constraints
8	AGE	0.2–1.0	High	5–20 years	Moderate	Ecological disruption
9	RCIPG	0.1–0.3	Moderate	2–5 years	Moderate	Retrofit costs
10	PAEMCS	0.1–0.5	Moderate	3–12 months	Moderate	Weather disruption

3.2. Chart: Potential Temperature Reduction by Method

Potential Global Temperature Reduction (°C)
2.0 |           █
1.5 |           █
1.0 |   █       █       █
    |   █       █       █       █       █
0.5 |   █       █       █       █       █       █
    |   █       █       █       █       █       █       █
0.0 |   █       █       █       █       █       █       █
    |   └───────┴───────┴───────┴───────┴───────┴───────┴─
      1       2       3       4       5       6       7

Legend:

• 1: PRCI

• 2: SRAI

• 3: SBRA

• 4: OSRF

• 5: HABF

• 6: TPS

• 7: CRCD

• 8: AGE

• 9: RCIPG

• 10: PAEMCS

(Note: For brevity, only first 7 methods shown in ASCII chart; see table for full data.)

3.3. Chart: Cost vs. Impact Analysis

Impact (High)
↑
|   SBRA  ●
|        ●
|   AGE  ●
|   ●   HABF
|   ●   SRAI
|   ●   PAEMCS
|   ●   TPS
|   ●   PRCI
|   ●   OSRF
|   ●   RCIPG
|   ●   CRCD
|────────────────────────→
Cost (Low)              High

3.4. Chart: Implementation Timeline

Time (Years)
0   5   10   15   20   25   30
|   |   |   |   |   |   |   |
PRCI    ██████████
SRAI    ██████
SBRA    █████████████████████
OSRF    ████████
HABF            ████████████
TPS             ████████
CRCD    ████
AGE                 ████████████
RCIPG       ████████
PAEMCS    ██████

---

4. Conclusion

Radiative cooling offers a diverse set of tools for geoengineering that can complement existing strategies for climate mitigation. The ten methods outlined above range from passive, low-cost surface coatings to complex, high-altitude and space-based systems. Each has distinct advantages and challenges, requiring careful consideration of scientific, economic, environmental, and political factors.

Key Findings:

• Passive radiative cooling surfaces (PRCI) and stratospheric radiative aerosols (SRAI) offer the most scalable and cost-effective near-term options.

• Space-based radiators (SBRA) and high-altitude balloons (HABF) provide higher potential impact but require significant investment and technological development.

• Oceanic films (OSRF) and cryogenic desalination (CRCD) offer dual benefits (e.g., water production) but are constrained by geography and ecological considerations.

• Artificial glacial expansion (AGE) and cloud-seeding modulation (PAEMCS) carry higher risks due to complex feedbacks and potential unintended consequences.

Recommendations:

• Pursue a mixed strategy combining multiple radiative cooling methods to balance risk and impact.

• Invest in research and development to improve material properties, reduce costs, and enhance scalability.

• Establish global monitoring and governance frameworks to oversee implementation and mitigate risks.

Radiative cooling represents a promising, underexplored frontier in geoengineering. With careful planning and international cooperation, it could play a critical role in mitigating the effects of global warming while providing ancillary benefits such as energy savings, water production, and infrastructure resilience.