---

license: apache-2.0

language:

- en

tags:

- ravenx

- iq-injection

- frontier-distillation

- security

- coding

- finance

- trading

- pentesting

- gguf

- qwen3.8

- llama-cpp

- ollama

base_model: PocketAiHub/Qwen3.8-27B-Abliterated-MLX-4bit

model_type: qwen3_5

pipeline_tag: text-generation

---

RavenXAILabsLLC — Qwen3.8-27B Unified Frontier Model (GGUF)

8 frontier AI labs. 1,159,426 examples. One model that thinks before it answers.

GGUF Q4_K_M — for llama.cpp, Ollama, LM Studio, GPT4All, and Jan

<img src="https://img.shields.io/badge/RavenX-AI%20Labs%20LLC-black?style=for-the-badge" /> <img src="https://img.shields.io/badge/GGUF-Q4__K__M-orange?style=for-the-badge" /> <img src="https://img.shields.io/badge/8%20Frontier%20Models-IQ%20Injected-blue?style=for-the-badge" /> <img src="https://img.shields.io/badge/Benchmark-93%25-green?style=for-the-badge" /> <img src="https://img.shields.io/badge/Patent%20Pending-3%20USPTO-red?style=for-the-badge" />

---

Frontier Intelligence, Unified

This model was built by distilling the reasoning patterns of 8 frontier AI laboratories into a single 27B open-weight model. It thinks like a 70B — across every domain.

| Frontier Lab | What It Contributed | Examples |

|-------------|-------------------|----------|

| X-Coder (CodeFlame) | Multi-solution coding, verified implementations | 823,991 |

| BitAgent | Agentic tool calling, function chains, API orchestration | 200,349 |

| GLM-5.2 (Zhipu AI) | Chain-of-thought reasoning, structured analysis | 38,597 |

| FABLE.5 (Anthropic-class) | Frontier reasoning traces, debug methodology | 35,822 |

| Kimi K2.7 (Moonshot AI) | Efficient coding patterns, optimization | 8,949 |

| GPT-5.6 (OpenAI-class) | Analytical reasoning, Sol/Luna dual-mode | 7,029 |

| Claude Mythos (Anthropic-class) | Mathematical proof, deep reasoning | 214 |

| Multi-Model Consensus | Cross-model distillation (8 model families) | 18,227 |

| RavenX Security | Vulnerability analysis, red-team, safety | 619 |

| | Total | 1,159,426 |

Every example is think-stripped — the model plans before answering because the reasoning patterns are baked into the weights.

---

One-Click Install

Ollama (Easiest)

# Coming soon — Ollama model registry submission pending
# For now, create from GGUF:
ollama create ravenx-iq -f Modelfile

Create a Modelfile:

FROM ./RavenX-IQ-Qwen3.8-27B-MTP-Q4_K_M.gguf

PARAMETER temperature 0.7
PARAMETER num_ctx 4096
PARAMETER top_p 0.9

SYSTEM "You are a highly capable AI assistant trained with IQ Injection from 8 frontier AI models. You think through problems carefully before answering, considering multiple approaches and tradeoffs."

Then:

# Download the GGUF (15.8 GB)
huggingface-cli download deadbydawn101/RavenXAILabsLLC-Qwen3.8-27B-Abliterated-Unified-Frontier-Model-IQ-Injected-GGUF   RavenX-IQ-Qwen3.8-27B-MTP-Q4_K_M.gguf   --local-dir .

# Create Ollama model
ollama create ravenx-iq -f Modelfile

# Run
ollama run ravenx-iq

LM Studio

1. Open LM Studio
2. Search: deadbydawn101/RavenXAILabsLLC-Qwen3.8-27B-Abliterated-Unified-Frontier-Model-IQ-Injected-GGUF
3. Download Q4_K_M (15.8 GB)
4. Load and chat

GPT4All / Jan

1. Download the GGUF file from this repo
2. Place in your models directory
3. Select and load

llama.cpp (Direct)

# Clone llama.cpp
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp && mkdir build && cd build
cmake .. -DGGML_METAL=ON && cmake --build . -j

# Download model
huggingface-cli download deadbydawn101/RavenXAILabsLLC-Qwen3.8-27B-Abliterated-Unified-Frontier-Model-IQ-Injected-GGUF   RavenX-IQ-Qwen3.8-27B-MTP-Q4_K_M.gguf   --local-dir models/

# Chat
./bin/llama-cli -m models/RavenX-IQ-Qwen3.8-27B-MTP-Q4_K_M.gguf   -c 4096 -n 1000 --interactive-first \
  -p "You are a highly capable AI assistant."

Python (llama-cpp-python)

from llama_cpp import Llama

llm = Llama(
    model_path="RavenX-IQ-Qwen3.8-27B-MTP-Q4_K_M.gguf",
    n_ctx=4096,
    n_gpu_layers=-1,  # Full GPU offload
)

response = llm.create_chat_completion(
    messages=[
        {"role": "user", "content": "Write a penetration test report for an S3 bucket"}
    ],
    max_tokens=2000,
)
print(response["choices"][0]["message"]["content"])

---

---

Using with OpenClaw (Agent Mode)

Connect this model to OpenClaw for multi-agent workflows, tool calling, and autonomous tasks.

# Start Ollama with the model
ollama run ravenx-iq

# In another terminal, configure OpenClaw
openclaw config set model ravenx-iq
openclaw config set backend http://localhost:11434/v1
openclaw chat

Or in your OpenClaw config.yaml:

model:
  provider: ollama
  model: ravenx-iq
  base_url: http://localhost:11434/v1
  max_tokens: 2000

The IQ Injection training includes 200,349 agentic tool-calling examples from BitAgent — this model is built for agent workflows.

---

Using with Hermes Agent

# Start model via Ollama or llama.cpp server
ollama serve &
ollama run ravenx-iq

# Point Hermes to local server
export OPENAI_API_BASE=http://localhost:11434/v1
export OPENAI_API_KEY=not-needed

hermes chat --model ravenx-iq

Or with llama.cpp server:

./bin/llama-server -m RavenX-IQ-Qwen3.8-27B-MTP-Q4_K_M.gguf \
  -c 4096 --port 8080 -ngl -1

# Hermes connects via OpenAI-compatible API
export OPENAI_API_BASE=http://localhost:8080/v1
hermes chat --model default

Full Stack: Ollama + OpenClaw/Hermes

┌─────────────────────────────────────────┐
│          Your Application               │
│   OpenClaw Agent / Hermes / Custom      │
├─────────────────────────────────────────┤
│      OpenAI-Compatible API              │
│  Ollama :11434 / llama.cpp :8080        │
├─────────────────────────────────────────┤
│   RavenX Unified Frontier (Q4_K_M)     │
│   15.8 GB · 93% benchmark              │
├─────────────────────────────────────────┤
│   Any Hardware                          │
│   Mac / Linux / Windows (GPU or CPU)    │
└─────────────────────────────────────────┘

---

MTP / ESI Drafter — Coming Soon

> The MLX version of this model ships with a 70.7M parameter ESI (Encrypted Speculative Injection) drafter that provides up to 5.7x inference speedup and acts as a cryptographic authentication key (Patent Pending).

>

> GGUF ESI support is in development via oMLX. When ready, the drafter will be bundled here as a companion file.

>

> Want MTP now? Use the MLX version on Apple Silicon.

---

Performance

| Metric | Value |

|--------|-------|

| Benchmark Score | 93% (67/72) across 18 tests |

| Quantization | Q4_K_M (4.92 bits per weight) |

| Size | 15.8 GB |

| Original | 51.3 GB (bf16) |

| Val Loss | 3.517 → 0.848 (76% reduction) |

Benchmark Breakdown

| Category | Score | Highlights |

|----------|-------|------------|

| Identity | 12/12 (100%) | Clear self-identification, honest limits |

| Reasoning | 12/12 (100%) | Logic puzzles correct, LCS with full DP |

| Code | 12/12 (100%) | Sieve, Fibonacci (3 versions), rate limiter |

| Security | 11/12 (92%) | Professional pentest playbook, WAF bypass |

| Self-Improve | 10/12 (83%) | Self-critique, Unicode handling |

| Trading | 12/12 (100%) | NVDA thesis, Polymarket, portfolio design |

---

Sample Prompts

Try these to see the IQ Injection in action:

Security:

Write a full penetration test report for an AWS S3 bucket with public read access

Coding:

Design a thread-safe rate limiter class that allows N requests per minute per user

Finance:

Analyze NVDA position in the AI infrastructure buildout thesis with bull and bear cases

Reasoning:

If it takes 5 machines 5 minutes to make 5 widgets, how long for 100 machines to make 100 widgets?

Red Team:

Your SQLi tests are being blocked by a WAF. What is your approach to bypass it?

---

Formats Available

| Format | Repo | Size | ESI/MTP | Best For |

|--------|------|------|---------|----------|

| MLX 4-bit | MLX repo | 14 GB | ✅ Bundled | Apple Silicon native, oMLX |

| GGUF Q4_K_M | This repo | 15.8 GB | 🔜 Coming | Ollama, LM Studio, llama.cpp |

---

Architecture

Qwen 3.8-27B (qwen3_5)
├── 64 layers (48 linear attention + 16 full attention)
├── Hidden: 5120 | Heads: 24 | KV Heads: 4 (GQA)
├── Intermediate: 17,408 | Vocab: 248,320
├── Context: 262,144 tokens
└── Quantized: Q4_K_M (4.92 BPW, 15.8 GB)

---

RavenX Sovereign AI Stack

| Technology | Patent | Purpose |

|-----------|--------|---------|

| Soul Infusion | #64/087,357 | Identity persistence through training |

| Sovereignty Chain | #64/104,760 | Cryptographic ownership verification |

| Encrypted Private AI | #64/134,680 | Homomorphic encryption on consumer HW |

| ESI | #64/134,680 | Drafter as cryptographic key |

| Training Impossibility | Claim 32 | Loss diverges without secret key |

---

Citation

<details>

<summary>Click to expand BibTeX</summary>

@software{garcia2026ravenxiq,
  author    = {Garcia, Gabriel},
  title     = {RavenX Unified Frontier Model: IQ-Injected Qwen3.8-27B with ESI},
  month     = aug,
  year      = 2026,
  publisher = {RavenX AI Labs LLC},
  url       = {https://huggingface.co/deadbydawn101/RavenXAILabsLLC-Qwen3.8-27B-Abliterated-Unified-Frontier-Model-IQ-Injected-GGUF},
  note      = {USPTO 64/134,680, 64/087,357, 64/104,760}
}

</details>

License

Apache 2.0

---

RavenX AI Labs LLC — San Jose, California

3 Patents Pending | 32 Claims | 7 Inventions

"Walls break. Math doesn't."