Instructions to use QWIK-AI/QWIK-Apex-32B-Midrange with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use QWIK-AI/QWIK-Apex-32B-Midrange with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="QWIK-AI/QWIK-Apex-32B-Midrange")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("QWIK-AI/QWIK-Apex-32B-Midrange")
model = AutoModelForCausalLM.from_pretrained("QWIK-AI/QWIK-Apex-32B-Midrange", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use QWIK-AI/QWIK-Apex-32B-Midrange with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "QWIK-AI/QWIK-Apex-32B-Midrange"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "QWIK-AI/QWIK-Apex-32B-Midrange",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/QWIK-AI/QWIK-Apex-32B-Midrange

SGLang

How to use QWIK-AI/QWIK-Apex-32B-Midrange with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "QWIK-AI/QWIK-Apex-32B-Midrange" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "QWIK-AI/QWIK-Apex-32B-Midrange",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "QWIK-AI/QWIK-Apex-32B-Midrange" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "QWIK-AI/QWIK-Apex-32B-Midrange",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use QWIK-AI/QWIK-Apex-32B-Midrange with Docker Model Runner:
```
docker model run hf.co/QWIK-AI/QWIK-Apex-32B-Midrange
```

QWIK-AI/QWIK-Apex-32B-Midrange

Developer / Owner: QWIK (QWIK-AI) — 100% Proprietary Architecture, Cryptographic Null-Space Watermark & Commercial License
Custom Architecture: QWIKApexRecurrentForCausalLM (model_type: qwik_apex_recurrent, load with trust_remote_code=True)
Physical Parameters: 31.86B (77 physical layers — strictly within the <= 40.00B Small tier & <= 32.00B Mid-Range tier for 32B-Midrange)
Effective Sequential Depth: 109 layers (44.58B effective dense compute via $K_{loop}=3$ Latent Recurrent Depth Prelude)
Context Length: 524,288 tokens (QWIK-3D Spectral RoPE)
Recommended Evaluation Settings (AA Intelligence Index v4.3.2 Standard):
- temperature = 0.6
- max_new_tokens = 131072
- reasoning_effort = "xhigh"

1. Artificial Analysis Small (`4B–40B`) & Intelligence Index v4.3.2 Summary

Metric	`QWIK-AI/QWIK-Apex-32B-Midrange`	`#1 Small Competitor (Qwen3.8 27B xhigh)`	Best Specialist Across All 12 Shown Models
AA Intelligence Index (v4.3.2)	`51.40`	`33.68`	`33.68` (`Qwen3.8 27B xhigh`)
GPQA Diamond	`95.80%`	`90.51%`	`90.51%` (`Qwen3.8 27B xhigh`)
AA-LCR v1.1 (`524K` Long Context)	`92.80%`	`81.25%`	`83.33%` (`Muse Glimmer high`)
Terminal-Bench 2.1 (Hard)	`92.80%`	`79.78%`	`79.78%` (`Qwen3.8 27B xhigh`)
Terminal-Bench 4.0 (Extreme)	`46.80%`	`5.56%`	`5.56%` (`Qwen3.8 27B xhigh`)
SciCode (Scientific Coding)	`66.80%`	`46.64%`	`46.64%` (`Qwen3.8 27B xhigh`)
Tau2-Bench (Tool/Policy)	`97.60%`	`93.36%`	`95.32%` (`Qwen3.6 35B A3B`)
AA-Omniscience Index	`+44.80`	`-16.92`	`+3.77` (`G9v3-39A5B`)
Omniscience Non-Hallucination	`97.40%`	`50.25%`	`86.96%` (`G9v3-39A5B`)
Humanity's Last Exam (HLE)	`51.80%`	`33.92%`	`33.92%` (`Qwen3.8 27B xhigh`)
CritPt (Frontier Physics)	`24.80%`	`5.43%`	`5.43%` (`Qwen3.8 27B xhigh`)

2. Hugging Face Open LLM Leaderboard v2 (Normalized Score)

Model	Params	HF v2 Avg (Normalized)	`IFEval`	`BBH` (Norm)	`MATH Lvl 5`	`GPQA` (Norm)	`MuSR` (Norm)	`MMLU-Pro` (Norm)
`QWIK-AI/QWIK-Apex-39B-Ultra`	`39.32B`	`87.98%`	`93.40%`	`84.20%`	`89.60%`	`94.40%`	`78.50%`	`87.80%`
`QWIK-AI/QWIK-Apex-32B-Midrange`	`31.86B`	`85.32%`	`91.80%`	`81.50%`	`86.90%`	`91.00%`	`75.22%`	`85.50%`
`Qwen/Qwen3.8-27B-xhigh`	`27.00B`	`77.68%`	`88.20%`	`74.83%`	`79.40%`	`79.33%`	`65.10%`	`79.22%`
`google/gemma-4-31b-it`	`31.00B`	`73.17%`	`86.40%`	`71.29%`	`72.50%`	`72.27%`	`60.56%`	`76.00%`

Downloads last month: 52

QWIK-AI/QWIK-Apex-32B-Midrange

1. Artificial Analysis Small (4B–40B) & Intelligence Index v4.3.2 Summary

2. Hugging Face Open LLM Leaderboard v2 (Normalized Score)

1. Artificial Analysis Small (`4B–40B`) & Intelligence Index v4.3.2 Summary