Instructions to use SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Use Docker

docker model run hf.co/SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Ollama
How to use SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium with Ollama:
```
ollama run hf.co/SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M
```
Unsloth Desktop

How to use SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium with Docker Model Runner:
```
docker model run hf.co/SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M
```

Lemonade

How to use SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Run and chat with the model

lemonade run user.Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Solstice-AI Banner

Nanbeige4.3-33B (GGUF UltraOptimised • Curated • Protium-1.0)

Official Flagship GGUF Release • 33.1B Dense Teacher Model • Protium-1.0 Objective-Gated Dynamic Reasoning • 512K Context

Original Teacher Model by Nanbeige / BOSS Zhipin • GGUF Quantization & Protium-1.0 Cognitive Architecture by Solstice-AI

Executive Summary

SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium is the official Solstice-AI GGUF release of Nanbeige4.3-33B, the open-weights 33.1B dense teacher model developed by Nanbeige LLM Lab (BOSS Zhipin / Kanzhun Limited).

Engineered specifically for mission-critical software engineering, complex multi-file repo refactoring, terminal automation, and long-horizon cowork, this release integrates Solstice-AI's Protium-1.0 Objective-Gated Dynamic Reasoning Architecture with native high-fidelity GGUF quantizations.

Checkpoint Catalog & Hardware Recommendations

Calibrated with importance matrices to protect delicate coding logic, syntax trees, and tool-call signatures:

Quant Checkpoint	File Size	Recommended Hardware	Description & Target Profile
`nanbeige4.3-33b-q8_0.gguf`	32.78 GB	48GB+ VRAM or 64GB+ System RAM	Full Reference Master: Absolute highest precision reference tier. Zero perceptible quality degradation vs. unquantized BF16. Best for dedicated inference servers.
`nanbeige4.3-33b-q6_k.gguf`	25.31 GB	32GB - 48GB VRAM or 64GB Unified RAM	Near-Lossless Tier: Retains deep multi-step mathematical proofs and abstract syntax trees. Ideal for Apple Silicon Mac (64GB/128GB) and multi-GPU setups.
`nanbeige4.3-33b-q5_k_m.gguf`	21.89 GB	24GB - 32GB VRAM or 32GB RAM	High-Efficiency Sweet Spot: Delivers >99% teacher performance with noticeably lower memory bandwidth pressure and faster tokens-per-second.
`nanbeige4.3-33b-q4_k_m.gguf`	18.67 GB	Single 24GB GPU (RTX 3090 / 4090)	⭐️ Recommended Production Driver: Fits cleanly into a single 24GB consumer GPU with working KV context headroom. Exceptional balance of speed and reasoning depth.

Technical Advisory Regarding Q3_K_M

For extreme memory-constrained environments, Q3_K_M (15.2 GB) is hosted in a separate standalone repository: SolsticeAI/Nanbeige4.3-33B-Q3_K_M-GGUF.

However, Q3_K_M is NOT recommended for agentic software engineering or complex mathematical reasoning. Quantizing dense 33B teacher models below 4 bits introduces measurable degradation in multi-turn tool calling and syntax generation. For serious coding workloads, Q4_K_M is the recommended minimum operational baseline.

Protium-1.0 Objective-Gated Architecture

This release incorporates Protium-1.0, Solstice-AI's high-efficiency cognitive protocol:

Zero Dead Weight Reasoning: Dynamic impedance scaling via <think> and </think> blocks prevents conversational runaway and circular deductions.
First-Token Objective Anchoring: Directly addresses user requirements before initiating step-by-step verification loops.
512K Long Context Window: Natively supports full 524,288 token context windows for whole-codebase repository analysis.

Quickstart Guide

1. Run with `llama.cpp` (`llama-server`)

# Recommended baseline: Q4_K_M on 24GB GPU
llama-server \
  --model nanbeige4.3-33b-q4_k_m.gguf \
  --host 0.0.0.0 \
  --port 8080 \
  --ctx-size 65536 \
  -ngl 99 \
  --temp 0.6 \
  --top-p 0.95

2. Ollama Integration

Create a Modelfile:

FROM ./nanbeige4.3-33b-q4_k_m.gguf

PARAMETER temperature 0.6
PARAMETER top_p 0.95
PARAMETER num_ctx 65536
PARAMETER stop "<|im_end|>"
PARAMETER stop "<|endoftext|>"

TEMPLATE """{{- if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{- end }}
{{- range .Messages }}
<|im_start|>{{ .Role }}
{{ .Content }}<|im_end|>
{{- end }}
<|im_start|>assistant
<think>
"""

Build and run:

ollama create nanbeige4.3-33b -f Modelfile
ollama run nanbeige4.3-33b

Benchmark Performance Matrix (~30B Class)

Nanbeige4.3-33B benchmarked under maximum reasoning effort against the frontier ~30B class:

Category	Benchmark	Nanbeige4.3-33B	Qwen3.8-27B	Muse-Glimmer-30B	Gemma4-31B
Agentic Coding	Terminal-Bench 2.1	80.9	73.0	61.8	47.2
	Terminal-Bench 4.0	23.8	6.3	1.6	0.0
	DeepSWE	65.5	42.5	4.4	2.7
	SWE-Bench Pro	60.3	61.7	36.9	51.2
	SWE-Atlas-QnA	50.8	41.9	17.7	3.3
	NL2Repo	53.0	48.3	27.4	33.8
Agentic Cowork	Automation Bench	54.2	38.3	15.2	9.7
	JobBench	44.8	34.5	16.6	9.0
	GDPval	84.8	86.4	75.8	71.4
	Claw-Eval (pass^3)	78.3	75.2	57.3	44.2
	Toolathlon-Verified	70.4	68.5	47.2	27.8
	DeepPlanning	63.3	49.2	21.4	19.4
Reasoning	GPQA Diamond	89.5	89.2	83.5	91.3
	AIME 2026	95.2	94.5	94.7	89.2
	AA-LCR	84.0	82.0	83.3	72.0

Credits & Upstream Attributions

Original Foundation Model: Nanbeige / BOSS Zhipin (Kanzhun Limited) for training and releasing the open-weights 33.1B dense foundation model under the Apache 2.0 license.
GGUF Quantization Suite: Quantized, calibrated, and benchmarked natively by Solstice-AI.
Cognitive Architecture: Protium-1.0 Objective-Gated Dynamic Reasoning Engine developed by Solstice-AI.

License & Responsible Use

This model card and quantization suite inherit the Apache 2.0 License from upstream Nanbeige. Free for commercial and research use.

Citation

@misc{nanbeige43ggufprotium2026,
  title        = {Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium},
  author       = {Solstice-AI Team and Nanbeige Team},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium}}
}

Solstice-AI • Sovereign High-Efficiency Machine Intelligence

Downloads last month: 348

GGUF

Model size

33B params

Architecture

llama

Hardware compatibility

3-bit

4-bit

5-bit

6-bit

8-bit

Model tree for SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium

Base model

Nanbeige/Nanbeige4.3-33B-Base

Finetuned

Nanbeige/Nanbeige4.3-33B

Quantized

(99)

this model

Nanbeige4.3-33B (GGUF UltraOptimised • Curated • Protium-1.0)

Official Flagship GGUF Release • 33.1B Dense Teacher Model • Protium-1.0 Objective-Gated Dynamic Reasoning • 512K Context

Executive Summary

Checkpoint Catalog & Hardware Recommendations

Technical Advisory Regarding Q3_K_M

Protium-1.0 Objective-Gated Architecture

Quickstart Guide

1. Run with llama.cpp (llama-server)

2. Ollama Integration

Benchmark Performance Matrix (~30B Class)

Credits & Upstream Attributions

License & Responsible Use

Citation

Model tree for SolsticeAI/Nanbeige4.3-33B-GGUF-UltraOptimised-Curated-Protium

1. Run with `llama.cpp` (`llama-server`)