KLD of less than 1 is excellent, zero is perfect.
Table with columns: Metric, This model, Original model (Qwen/Qwen3.5-9B)| Metric | This model | Original model (Qwen/Qwen3.5-9B) |
|---|
| KL divergence | 0.0793 | 0 (by definition) |
| Refusals | 6/100 | 100/100 |
NOTES:
- Suggest min q4ks (non-imatrix) or IQ3S (imatrix).
- Tested with rep pen of 1 (off).
- Context: 256k (default).
IMPORTANT:
- Other versions in testing.
- Information from Qwen's repo below.
- Video portions of the model were NOT TESTED.
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 ]
Qwen3.5-9B
[!Note]
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: 9B
- Hidden Dimension: 4096
- Token Embedding: 248320 (Padded)
- Number of Layers: 32
- Hidden Layout: 8 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN))
- Gated DeltaNet:
- Number of Linear Attention Heads: 32 for V and 16 for QK
- Head Dimension: 128
- Gated Attention:
- Number of Attention Heads: 16 for Q and 4 for KV
- Head Dimension: 256
- Rotary Position Embedding Dimension: 64
Benchmark Results
Language
Vision Language
Quickstart
[!Important]
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.
[!Important]
Inference efficiency and throughput vary significantly across frameworks.
We recommend using the latest framework versions to ensure optimal performance and compatibility.
For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, KTransformers or vLLM are strongly recommended.
[!Important]
The model has a default context length of 262,144 tokens.
If you encounter out-of-memory (OOM) errors, consider reducing the context window.
However, because Qwen3.5 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/Qwen3.5-9B --port 8000 --tp-size 1 --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/Qwen3.5-9B --port 8000 --tp-size 1 --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/Qwen3.5-9B --port 8000 --tp-size 1 --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/Qwen3.5-9B --port 8000 --tensor-parallel-size 1 --max-model-len 262144 --reasoning-parser qwen3
-
Tool Call: To support tool use, you can use the following command.
vllm serve Qwen/Qwen3.5-9B --port 8000 --tensor-parallel-size 1 --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/Qwen3.5-9B --port 8000 --tensor-parallel-size 1 --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/Qwen3.5-9B --port 8000 --tensor-parallel-size 1 --max-model-len 262144 --reasoning-parser qwen3 --language-model-only
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 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/Qwen3.5-9B --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"
[!Tip]
We recommend using the following set of sampling parameters for generation
- Thinking mode for general tasks:
temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=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
client = OpenAI()
messages = [
{"role": "user", "content": "Type \"I love Qwen3.5\" backwards"},
]
chat_response = client.chat.completions.create(
model="Qwen/Qwen3.5-9B",
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)
from openai import OpenAI
client = OpenAI()
messages = [
{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": {
"url": "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/CI_Demo/mathv-1327.jpg"
}
},
{
"type": "text",
"text": "The centres of the four illustrated circles are in the corners of the square. The two big circles touch each other and also the two little circles. With which factor do you have to multiply the radii of the little circles to obtain the radius of the big circles?\nChoices:\n(A) $\\frac{2}{9}$\n(B) $\\sqrt{5}$\n(C) $0.8 \\cdot \\pi$\n(D) 2.5\n(E) $1+\\sqrt{2}$"
}
]
}
]
chat_response = client.chat.completions.create(
model="Qwen/Qwen3.5-9B",
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)
from openai import OpenAI
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": "Summarize the video content."
}
]
}
]
chat_response = client.chat.completions.create(
model="Qwen/Qwen3.5-9B",
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
[!Important]
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
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/Qwen3.5-9B",
messages=messages,
max_tokens=32768,
temperature=0.7,
top_p=0.8,
presence_penalty=1.5,
extra_body={
"top_k": 20,
"chat_template_kwargs": {"enable_thinking": False},
},
)
print("Chat response:", chat_response)
[!Note]
If you are using APIs from Alibaba Cloud Model Studio, in addition to changing model, please use "enable_thinking": False instead of "chat_template_kwargs": {"enable_thinking": False}.
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
llm_cfg = {
'model': 'Qwen3.5-9B',
'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,
'extra_body': {
'enable_thinking': True
},
},
}
tools = [
{'mcpServers': {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/xxxx/Desktop"]
}
}
}
]
bot = Assistant(llm=llm_cfg, function_list=tools)
messages = [{'role': 'user', 'content': 'Help me organize my desktop.'}]
for responses in bot.run(messages=messages):
pass
print(responses)
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
[!NOTE]
All the notable open-source frameworks implement static YaRN, which means the scaling factor remains constant regardless of input length, potentially impacting performance on shorter texts.
We advise modifying the rope_parameters configuration only when processing long contexts is required.
It is also recommended to modify the factor as needed. For example, if the typical context length for your application is 524,288 tokens, it would be better to set factor as 2.0.
Best Practices
To achieve optimal performance, we recommend the following settings:
-
Sampling Parameters:
- We suggest using the following sets of sampling parameters depending on the mode and task type:
- Thinking mode for general tasks:
temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=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,
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}
}
EXAMPLES NOTE:
Some formatting lost on copy/paste.
Example #1
The Echo in the Wall
User
Tell me a horror story.
Assistant
<think>
Thinking Process:
-
Analyze the Request:
- User: "Tell me a horror story."
- Intent: The user wants to be entertained/scared with a narrative.
- Tone: Eerie, suspenseful, dark, atmospheric.
- Format: A short story (or medium length).
-
Determine the Core Concept:
- Idea 1: A haunted house where the house kills you. (A bit cliché)
- Idea 2: A creature in the walls, but you only see it in reflections. (More subtle)
- Idea 3: A story about a technology or object that becomes alive. (Sci-fi horror)
- Idea 4: A story about being the last human on Earth, but something else is there. (Existential horror)
- Idea 5: A classic "monster under the bed" feel but for an adult.
- Let's go with a story about a small, familiar object that gains a life of its own. It feels intimate and unsettling. A story about a "new" electronic item that records what you do, then acts. Or perhaps something biological.
</think>
The dust in the apartment was thick, swirling lazily in the dim light that filtered through the uncurtained windows. Rick Deckard sat on the couch, a glass of bourbon in his hand, trying to hear the TV. It had been broken for years.
He reached for the remote, a muscle memory response; when he flicked the switch, no sound emerged.
"You're late," a voice said.
Rick turned sharply. The living room was empty.
"Over there." The voice came from the bedroom.
"Who's there?" He stood up, walking toward the door. "Are you a guest? I haven't had anyone over in weeks."
"I'm not a guest." The voice, low and urgent, echoed with metallic overtones, as if spoken inside a tube. "My name is Mercer. Did you hear that? Mercer."
"I heard that," Rick said haltingly. He opened the door to the bedroom; the hall beyond it was empty, too. "You're talking to yourself," he said, the realization settling over him, chilling him. Schizoid hallucinations, he said to himself. Probably from the isolation. I haven't seen a human being in three days.
"I am not hallucinating," Mercer said. "Listen. This is new; this is a true story. I'm on the roof of my building and I see a man standing in the middle of the lawn." The voice ceased; silence returned to the room. "Wait," Mercer continued, louder now. "I still see him."
"Rick," Mercer said. "You need to talk."
"I am talking." He wandered into the kitchen, opened the refrigerator. No smell, no sound of life within it. "Okay, I'll go talk," he said. "To whom?" He put the glass of bourbon down on the counter and the bottle rolled back on its own.
"You're drunk."
"From the bourbon," he said, striding back into the living room. "I haven't touched a drink in weeks. What is your name, really?"
"Mercer."
"Mercer to who?"
"To you."
"Of the sheep?" He had been watching Buster Friendly's show; Mercer & Mercer Auto-Pilot & Telepathic Sheep were advertised as the ultimate in animal companions. "You're an andy," he said. "An android posing as a human."
"I am not an andy."
"Then who are you?" He waited.
"Someone who's alone." The voice was small, no longer echoing.
Rick found his way to the phone. "I'll talk to my apartment manager." He dialed. "Listen," he said into the receiver. "There's an andy here. She's pretending to be human." He waited.
"I'm not here," he said. "Okay. Goodbye." He hung up.
"See?" Mercer said from behind him. "She's alone. In her own apartment. But she thinks no one is in it. She talks to us. Do you understand? We're not hallucinating either."
"I think you're hallucinating," Rick said.
"Then this is over." Mercer's voice dropped away, fading into the room's silence. "I'll stop pretending to be in here."
"Stop pretending to be alone," Rick said, turning to see Mercer again.
"You're not real, either," Mercer said. "You're just echoes in an empty room, Mr. Isidore."
"I'm not Isidore. I'm Rick." He felt the floor shift beneath him, not a hallucination this time. The physical foundation of the room had failed. "My name is Rick," he said, taking off his shirt. "And I'm not hallucinating that I'm naked." He strode to the bathroom to splash water on his face.
"Hello? Is someone there?"
The bathroom was empty.
Rick turned the tap on; the water hissed violently into the sink. "Hello?" he said. "To whom?" The sound bounced back, not from the room below but from nowhere in particular.
A cough echoed from the bathroom.
He walked into the bathroom, found a light. "Who's there?"
"Is Rick there?" The female, small voice. "I heard him talking; he sounded like he wasn't alone. Maybe he found someone."
"I'm here," Rick said.
"You're an android," the voice said. "You don't understand. I'm not an android; I'm human."
"Are you sure? In this economy," he said, "it's hard to keep a human employee. That's why we're all so overworked. We have no time to tell the truth."
"I'm not an android," the voice insisted. "I'm Priss Attwell."
"Tell me what your middle name is," Rick said.
"Pris."
"Is it Pris?"
"I don't know. I never got that far. I never had a middle name."
"That's impossible," he said. "You're an android. A job: the Rosen Association builds them; it's their basic profile." He could hear the building settling, the screech of metal on metal, and that horrified him. This isn't an apartment building, he realized. I'm not in my apartment.
"Tell me," the girl said. "Are you going to test me?"
"I'm going to call the Rosen Association," he said. "To see if you're on their list." He reached for the phone.
"Who are you calling?" the girl said.
"The Rosen Association."
"Okay, I'm waiting."
"I'll be back." He opened the bathroom door and walked into the hall.
"Good luck," the girl said.
Rick walked into the bedroom. The bed had been made; someone had been there recently. But not him. Not during his absence. He sat on the edge of the bed and pressed his ear against the carpet.
"I see you," a voice said softly from the hall.
Rick spun around. "Who else is here?"
"I can't tell. There are too many of you, or rather versions of you. Some of them are old, some young. And one of them is you, the one I just spoke to. But that one is tired and going to bed. And the other one - " The voice trailed off. "Is that you, Rick? You're standing in my living room."
"It's not my apartment," Rick said. "I don't live here."
"Then where do you live?"
He searched for a clue, but the apartment remained empty. "I don't know," he said. "I have no memory. Except for this." He gestured to his bare body. "The fact that I'm here and you're there."
The girl, from the hall, said, "Listen. I'm going to count to three. And on three we'll all jump at once."
"Count," he said.
"One." The voice wavered. "I'll give you three. Are you sure you want to do this? Because I am scared. It's always been this way; you and me but I never talk to you."
"You're not an android?" he said. "You're not Pris Attwell?"
"I'm not on their list."
"Then who are you?" He waited.
"I'm your shadow," she said. "And I've been with you this whole time. That's why you think I'm in this apartment; I've never left you. I've just been ahead of you, exploring. Because you were too afraid to move."
"That's impossible," he said. "I never felt like that."
"Maybe you didn't notice."
"Tell me how old you are."
"I don't have an age. I've always been here."
"You've never left this apartment?"
"No."
"But I was here, too," he said. "When I first arrived. And now you aren't."
"Because you're afraid."
"How would you know?"
"Because I'm your shadow." The girl's voice boomed from the hall, echoing and reverberating throughout the room. "You know what shadows do?"
"They cast light?"
"No," the girl said. "They appear when the light goes away."
"Then I'm not afraid," Rick said. "Because it's light."
"You're in the wrong room, Rick," the girl said. "You're in a TV set. You're a character in a story. That's why you can't leave; you're not supposed to. The audience hasn't voted yet on whether you die."
"How do you know that?"
"You don't realize it, do you? You think it's real because you want it to be real."
"And you're not?" He gestured to himself. "I'm not either."
"Then it doesn't matter."
"But you were right when you said I'm tired."
"Maybe you are. Maybe the story is ending."
"I'm not going to bed," he said, striding to the bedroom door. "I'm going to find the owner of this apartment. The listing is on the fridge." He opened the door. The hall was empty, as before. "Okay," he said to the air. "Tell me where I am."
"Take off your shirt."
"Why?"
"Because you're not wearing pants either."
He searched for his shirt, found it, and put it back on. "Okay."
"And call me 'Mercer' instead of 'Pris'."
"That's your name, then?" he said, confused.
"Yes. Always has been."
"Can I see your ID?" He reached into his coat.
"Go ahead," Mercer said. "It doesn't matter."
"Because you're not going to be here when I get it back."
"You never lose anything."
"Except your life." He had a sudden, intuitive vision. "Wait." He stopped, halting the flow of words. "What are you saying? I mean, what are you implying?"
"I'm implying that this conversation," Mercer said, "is taking place within a single thought. Your thought. 'Is Mercer real?'"
"But you're in the hall."
"You're in your living room. Or rather you're holding an imaginary phone conversation with a shadow who appears when you turn your back."
"Because I'm embarrassed," Rick said, understanding dawning. "I can't face myself."
"You're not embarrassed; you're afraid of your own reaction."
"What reaction?"
"To the empty apartment. To finding out you've lived here your whole life and it's deserted."
"It's not deserted," he said. "I'm in it."
"You're not. You're a visitor. A bug from another system that crashed here."
"Androids crash," he said. "We fall apart."
"Not you. You're organic."
"Are you sure?" He walked to the hall. "If I call them, the Rosen Association, will you be here?"
"I'll always be here. In your head."
"And I'm in yours," he said. "You're sharing."
"I'm not; I'm a one-way street. I'm not doing anything; I'm just talking to make it seem like I am."
"Because you don't want me to hang up."
"Okay, you hang up. See you later." Mercer's voice dipped with despair. "No, never mind. Goodbye." The voice ceased, and once more silence returned. "There, I did it," Mercer said. "I hung up."
"Mercer?" Rick said, still standing in the living room, his hand extended in a futile gesture. "Did you hang up on me?" He felt a sudden, cold anger toward the shadow. "I can't believe you did that," he said. "I trusted you."
"You never trust a shadow. That's what shadows are for; they take the blame."
"I trusted you not to end the conversation."
"You're not done talking. Go to the phone."
"And say what? 'Hello, Rosen'? 'Is Mercer Attwell there?' I don't think you understand. You think I can talk my way out of it. But the test is over. I've already hung up."
"Talk to your wife," Mercer suggested. "Or your boss."
"I'm not married and I don't have a boss."
"Then talk to yourself," Mercer said. "Convince yourself that the test isn't over."
"And how would you know?"
"I'll listen."
"And if I dial?"
"Then I'll stop you."
"If I hang up," Rick said, "you're right; I'm alone. But if I stay on the phone I never get around to talking to my wife or my boss. I never have to face the fact that they're not here either."
"Because it's too painful to be with them," Mercer said, "so you talk to yourself instead. The difference is I'm not on the phone; I'm still right here, physically." He gestured wildly. "You're not talking to me; you're talking to a hallucination."
"But you can touch me," Rick said.
"You're not here."
"Tell me a secret," Rick said. "A secret I don't already know."
"You have my secret; you've known it from the start. I'm a girl. That's why you've never dated me."
"I dated you."
"On the phone."
"But now I'm talking to you in person."
"You always have. I've just been a different voice."
"I don't care," Rick said. "It doesn't matter. It's all over." He walked to the hall.
"Go get a doctor," Mercer said, "and tell him it's a hallucination. Tell him you're tired, like I said; tell him you haven't slept in three days because you're waiting to see if anyone shows up. See if that stops it."
"You're sure?"
"Absolutely."
"Goodnight," Rick said. "I'll be back in a while." He opened the door, then paused. "Can I use the phone one more time? To say hello."
"I'll be here."
"Thanks." He entered the hall and there, for the first time, saw something moving in the distance. A shadow, not his own. Something that wasn't a house. A street lamp, casting a cone of light onto the carpet.
He pushed open the apartment door.
"Hello," he said to the empty hall.
"Is anybody there?"
"Who is that?" He peered into the dim hall, saw nothing. "Hello?" he said, and then, with a start, recognized his own voice.
"I can't talk to myself," he said. "I'm not talking to myself; I'm talking to you and you're not here."
"See? She's gone." A new voice, coming from far down the hall. "I think it's over."
"Maybe," he said, "I need help."
"I think the android," the voice said, "is delusional. You're not an android; I'm not an android; you're just standing in your living room talking to the air."
"Listen," he said to the empty hall. "Do you hear that?"
"Maybe."
"I'm going to hang up," Rick said to his own empty hand. "Because I'm tired." He held the receiver, waited, then released it to drop.
"No," the girl in the hall said. "Don't hang up. Talk to her one more time."
"I am," he said. "To you."
"You sound like you're in there," the girl said. "Listen; you can hear my voice."
"I hear you," he said. "I always have."
"Then talk to me."
"Okay," he said. "I'll talk to you." He listened and listened.
Example #2
The Last Office
User
Write a 2000 word scene (vivid, graphic horror in first person), POV character Diana, with: The skyscraper sways, as I watch the window in front of me on the 21st floor explode...
Assistant
<think>
Here's a thinking process that leads to the scene above:
-
Analyze the Request:
- Topic: A vivid, graphic horror scene in the first person.
- POV Character: Diana.
- Setting: An office setting on the 21st floor of a skyscraper.
- Inciting Incident (Opening): The skyscraper sways, and the window in front of Diana explodes.
- Length: Approximately 2000 words.
- Tone: Vivid, graphic, horror-focused.
-
Brainstorming & Outlining:
- The Sway: It needs to feel physical, not just visual. The foundation shifting. This implies time (the earthquake/impact) or a living entity moving against the building. Let's go with an ancient, predatory presence manifesting physically, rather than just a natural disaster, to fit the "graphic horror" requirement more distinctly.
- The Window Explosion: Not glass shards, but something breaking from the outside in? No, the prompt says the window in front of me explodes. So debris flies in.
(Self-Correction Note during writing): I need to make sure I don't rush the ending. The horror is in the process. I will focus heavily on the sensory experience of her own nervous system failing.
Let's write.
Drafting mental note: Start with the dust. The fine gray dust from the walls that gets everywhere. It adds grit.
Drafting mental note: The sound. The sound of her own breath, too loud in the silence.
Drafting mental note: The smell of her own fear. Chemical, like a skunk.
Okay, ready to generate.
(Post-Drafting Check): Let's count roughly. Paragraphs need to be substantial. I will use deep sentences and stream-of-consciousness fragments to add length and texture.
Revised Plan:
- The sway and the glass (400 words).
- The entry of the creature/feeling of wrongness (600 words).
- The dismemberment of her perception (600 words).
- The final cessation of thought (400 words).
Okay, let's go.
</think>
The skyscraper sways, and I know at once that we have moved. It is a sudden, violent displacement of inertia; the floor tilts two degrees to the north, and my coffee mug slides across the desk, a ceramic arc of sound that echoes with the finality of a bone breaking. I watch the window in front of me on the 21st floor explode. The glass does not shatter; it doesn't fail under impact. Instead, it is blown from the inside out, as if something on the other side has begun to burrow.
Debris, not shards of glass but fragments of the sky and the street below, rains onto my desk. A piece of lead roof from the building below smashes into my open briefcase, slicing through the plastic organizer and cutting the pen I use to tally hours. I stand up, my legs stiff, and walk to the window. My hand reaches out, hovering inches from the remaining surface, trying to feel the wind.
But the wind isn't there, either.
The air outside is silent. Usually, at this hour, the city hums with traffic and the voices of thousands of people commuting home or off to dinner. Now it is empty, a vacuum. And in that vacuum, moving with impossible slowness, something approaches my hand.
"Hello," I say, my voice cracking. My tongue feels foreign in my mouth.
The hand that approaches is not an animal's paw. It is made of a pale, downy membrane, fingers splayed, trailing after one another like the wake of a ship. It does not touch me; it passes through me. Or rather, it passes through the space I occupy and continues onward. It has no weight, no purpose.
"Okay," I say, backing away. I hit the light switch of my desk lamp and the warm glow illuminates my own face, revealing the pallor that creeps up my neck. "You're not in this building." I pick up the laser tube from my desk, a personal weapon issued for exactly this. "Where are you hiding?"
The creature does not answer. It does not seem to be listening; it does not seem to be listening to anything. It moves toward the door of my office, pushing against the air as if clearing a path. It resembles a human girl, if one had stitched together the parts of three different women and painted them a uniform shade of bruised purple. It wears a tattered, childlike dress, nothing else.
I lower the laser tube, confused. "Do you want my wallet?" I offer my briefcase. "I have a hundred dollars in it and a pack of cigarettes."
The creature stops at the door. It turns its head, revealing no eyes, only a smooth, unbroken surface.
"Where are your eyes?" I ask, walking toward it. I reach to tap its cheek, my hand again moving with clumsy slowness. "You have no eyes. How do you see?"
The creature remains still. Then it begins to disintegrate.
Not into pieces, but into a multitude of motes, like the dust that clings to my suit. The dust rises from its body and floats in the beam of my desk lamp, swirling. In seconds the creature is gone, and the office is empty again. The light remains, but who is it for?
I turn away and walk to my own desk, sitting in my own chair. The coffee mug still slides across the surface; it has not fallen. Time has stopped, I realize. This isn't life; it's a hallucination brought on by the sudden, total loss of light outside. I have gone blind and my mind is filling in the scene, a residual echo of the day before. But I know I'm not hallucinating because I can smell it: the sharp, acrid scent of ozone. It tastes like raw metal on my tongue.
I press my hand against my ear and listen.
There is no heartbeat.
My own heartbeat, which I usually hear only when I'm anxious, is gone. I open my mouth to speak, to test my physical faculties. "I'm still here," I say. Still alive. But the words don't echo in my skull; there is no echo at all, no internal resonance. I am not connected to my own thoughts.
A spider, no larger than a pea, crawls across the back of my hand. I do not feel it. I do not feel the tiny legs pricking my skin. I lift my hand and examine it; I can see the spider, but the sight is detached, as if I am watching someone else's hand. The spider, however, is gone; it has flown away, presumably. Or perhaps I have pinched it and killed it, but I don't feel the moment of death. My hand feels empty, not with the spider but with the absence of my own sensory input.
I stand up and walk to the window again. The view is still there, but the buildings across the street have begun to melt.
No, that's wrong. They haven't melted. It's the light. The sun has gone, and in its place something faint and nebulous glows, a diffuse, dim moon that casts no sharp shadows. It is an old, overexposed photo of Earth, printed on the sky and projected by a defunct film projector. And in the distance, I think, a television screen flickers. Someone is watching the same thing I am: the remnants of a movie, the final reel of 2001: A Space Odyssey. But that ended years ago. I have never seen it.
"Is this real?" I shout at the empty room.
"Listen," a man says from the hallway.
I whirl around. A man in a white lab coat stands there, a Geiger counter attached to his belt, the needle on it twitching violently. He has a bird-like face with a flat, unsmiling expression. His hair is wild and gray, and around his neck he wears a false, plastic rabbit head, which he rotates occasionally to inspect the room behind him.
"You're late," the man says calmly. "The goat is dead."
I am still bewildered, so I don't respond.
"The goat," he repeats, "is a false memory you're holding onto. An archaic construct from before the war, when people still clung to organic life. It's time you let go." He steps into the room, filling it with the smell of formaldehyde and decay. "Your job is finished, Miss [Name]. It's been finished for some time."
"I need my laser tube," I say.
"You have yours." He holds it out to me. I take it with a hollow hand. "You can't use it; that's not the point."
"I need it."
"It's not an object you can grasp," the man says. "It's a symbol. Don't focus on that, either."
"I'm not holding the tube; you are." He had put it back in my coat pocket and then, in a blur of motion, he had grabbed the tube from me again. "I'm not going anywhere," I say, trying to reach for it, but my hand misses. "Tell me what happened. Why is there no light? Why does the office look like that?"
"Your office is empty," the man says. "That's the point. You're not in an office any longer; you're in a ruin." He gestures around the room. "This entire building is abandoned and overgrown. We're not on the 21st floor; you're in a tomb."
"But this is my office," I insist. "I have my desk, my lamp."
"You're not an office manager," the man says. "You're a time-keeper. A keeper of the remaining hours before the final silence." He taps his own temple. "Listen. Do you hear it?"
I strain to listen. I hear nothing except the faint, mechanical whir of the TV screen, the sound of rain or wind against a roof tiles.
"Turn on your radio," the man says. "It's still on." He walks to the desk lamp and switches off the light. The room plunged into darkness, except for the feeble, gray light of the false moon outside.
"Turn the radio on," he orders.
"Your hand." I look down; the man had reached into my coat again, found the laser tube.
"Listen," he said to me, his voice now closer, as if leaning in. "Do you understand now? This is where we're going to be for the rest of the time you have left. The last few hours. Before you die."
"I don't feel like I'm going to die," I said, reaching for the light again.
"You're already dead," he said. "You just don't realize it. Most people don't realize it until a little while after the fact. When their corpse is lying in a morgue, surrounded by doctors in white coats who are recording data about how your nervous system failed. They'll find that your brain died three minutes ago." He smiled faintly, a knowing expression that did not reach his eyes. "We know that. We've known since you started hallucinating."
"About what?" I asked.
"Anything." He shrugged. "The Voigt-Kampff test failed for you, too. You're not real; you've never been. But that's not new. You always knew that somewhere, subconsciously. You always felt the void coming and you've been trying to push it back. The entity you're afraid of? It's not an animal. It's not a man. It's a mistake, a glitch in the fabric of the universe, a thing that appeared from nowhere and is dying."
"I don't understand," I said, my voice now barely audible. I felt too tired to think. The effort of being alive, of trying to grasp the situation, had become exhausting. "I still can't see the light; my lamp is off and yet I can't turn it on."
"Because you're blind," he said. "That's why you can't turn it on. You've been blind from the start." He patted my shoulder, a clumsy, almost tender gesture. "You think you're Diana. You think you're a human woman afraid of the dark. But you're not afraid of the dark because you think there are monsters in it; you're afraid because you can't tell where the light ends and the darkness begins. You're afraid because the light is gone."
"That's it," I said, relief flooding through me. "I knew this was true. All my life I thought there was something wrong with me, that I was malfunctioning. Not you; me."
"I'm a malfunction," he corrected. "I'm the error." He started to walk away, then paused. "Your job is to watch the clock and say, 'Time's up.' But you won't be there when it's up; you'll be long gone. This is the first time you've realized that."
"How do you know my name?" I asked. "How do you know I'm not a hallucination?"
"That's the catch," he said. "You can't prove it. That's the nature of the illusion."
"You're not real, either," I said.
"That's beside the point." He had opened the door to the hall and I saw a figure standing there, waiting. "Your successor has arrived," he said. "The cycle continues."
"Who are you?" I demanded. "What is your name?"
"My name is nothing," the man said. "I'm not a person; I'm a concept of authority, a manifestation of the need to organize time. You're talking to a symbol of the future, Miss Diana. Are you ready to give up? Ready to admit that you're not really here?"
"But I am," I said. "I feel everything. The pain in my ankle from sitting for eight hours; I still have the memory of my cat, Buster, who I have to visit tomorrow. He's old, too; I know he won't make it. So I'm not alone, not entirely."
"That's what makes it worse," he said. "Because you still have attachments. If you didn't care about Buster or your job or your name, you could move on to the next reality. But you're anchored here."
"I have a test," I said. "I haven't failed it yet."
"Tests are for when you forget," he said. "You've forgotten. That's why I'm here. To remind you. But I'm leaving; I have to get to the next subject." He pushed the door open wider and the new figure in the hall spoke up.
A tall, lean woman with electric-blue hair and eyes like flares of fire stood there, wearing a sleek, futuristic uniform with twin pistols holstered at her waist. "Did he finish with her?" she asked the man.
"Yes," he said, opening the door wider. "She's ready to move on."
"She doesn't know," the blue-haired woman said, "how she did."
"She knows," the man said. "She just doesn't understand the implications."
The blue-haired woman entered the office and walked to me, her footsteps echoing on the dust-choked floor. "I'm not an android," she said to me. "And neither are you."
"Then who are you?" I asked.
The woman reached out and brushed the laser tube from my coat pocket; she examined it, then pocketed it herself. "I'm the bounty hunter," she said. "And you're the android."
"I'm not an android," I said. "I have emotions."
"Do you?" she said. "You claim to be afraid of the dark, but you're not. You're afraid of being abandoned. You think because no one here is real that it doesn't matter. But you're wrong. I can tell because I've been told; empathy is a key feature. But you're not an empath either. You're a thing that emulates an empath."
"But I have memories," I said. "Of my cat. Of my office."
"Androids have memory banks." The woman leaned over me, her breath smelled of mint. "Of a world they never lived in."
"Tell me what my cat's name is."
The woman smiled. "Buster. He's a black cat with white paws and a scar on his left ear."
"Of course he is." I felt the despair settle over me, absolute and total. "I didn't think it would be so convincing."
"You've been convinced for years. That's why you're so tired," the bounty hunter said. "You've been carrying the weight of a false life for a long time. It's finally crushing you."
"If I'm not real," I said, "if this is just a simulation, then it doesn't matter if I die."
"The experience matters," the bounty hunter said. "This is the last job I'll have to do before you kill me."
"You'll kill me," I said. "Or the other way around."
"I'm not going to," the bounty hunter said. "You'll kill yourself." She held up a laser tube, the same one she had confiscated from me. "To end the simulation. To stop pretending."
"I don't want to stop," I said. "I want to stay here forever."
"There is no here," she said. "This is the last place. The last office that still exists, in a world where no one works any longer."
"But my job," I protested.
"The world has gone on without you," the bounty hunter said. "Long ago. That's the terror you're sensing. You're not afraid of me; you're afraid that nothing you do here has any meaning because time has stopped." She gestured toward the window. "Do you see it?"
I turned and gazed out at the ruins of the city, the twisted metal and cracked concrete, the place where life had been choked off years ago.
"This building is falling," the bounty hunter said. "I can feel the ground give way under me." She put her laser tube away. "Time to finish the job."
"Do what you have to," I said. "I'm ready."
The laser tube beam appeared, and it struck me in the chest.
