aquif-3.5-Max-42B-A3B GGUF Models
Model Generation Details
This model was generated using llama.cpp at commit 7f09a680a.
Quantization Beyond the IMatrix
I've been experimenting with a new quantization approach that selectively elevates the precision of key layers beyond what the default IMatrix configuration provides.
In my testing, standard IMatrix quantization underperforms at lower bit depths, especially with Mixture of Experts (MoE) models. To address this, I'm using the --tensor-type option in llama.cpp to manually "bump" important layers to higher precision. You can see the implementation here:
👉 Layer bumping with llama.cpp
While this does increase model file size, it significantly improves precision for a given quantization level.
I'd love your feedback—have you tried this? How does it perform for you?
Click here to get info on choosing the right GGUF model format
aquif-3.5-Plus & aquif-3.5-Max
The pinnacle of the aquif-3.5 series, released November 3rd, 2025. These models bring advanced reasoning capabilities and unprecedented context windows to achieve state-of-the-art performance for their respective categories.
aquif-3.5-Plus combines hybrid reasoning with interchangeable thinking modes, offering flexibility for both speed-optimized and reasoning-intensive applications.
aquif-3.5-Max represents frontier model capabilities with reasoning-only architecture, delivering exceptional performance across all benchmark categories.
Model Repository Links
| Model | HuggingFace Repository |
|---|---|
| aquif-3.5-Plus | aquiffoo/aquif-3.5-Plus |
| aquif-3.5-Max | aquiffoo/aquif-3.5-Max |
Model Overview
| Model | Total (B) | Active Params (B) | Reasoning | Context Window | Thinking Modes |
|---|---|---|---|---|---|
| aquif-3.5-Plus | 30.5 | 3.3 | ✅ Hybrid | 1M | ✅ Interchangeable |
| aquif-3.5-Max | 42.4 | 3.3 | ✅ Reasoning-Only | 1M | Reasoning-Only |
Model Details
aquif-3.5-Plus (Hybrid Reasoning with Interchangeable Modes)
A breakthrough hybrid reasoning model offering unprecedented flexibility. Toggle between thinking and non-thinking modes to optimize for your specific use case—maintain reasoning capabilities when needed, or prioritize speed for time-sensitive applications.
Artificial Analysis Intelligence Index (AAII) Benchmarks
Core Performance Metrics
| Benchmark | aquif-3.5-Plus (Non-Reasoning) | aquif-3.5-Plus (Reasoning) | aquif-3.5-Max |
|---|---|---|---|
| MMLU-Pro | 80.2 | 82.8 | 85.4 |
| GPQA Diamond | 72.1 | 79.7 | 83.2 |
| AIME 2025 | 64.7 | 90.3 | 94.6 |
| LiveCodeBench | 50.5 | 76.4 | 81.6 |
| Humanity's Last Exam | 4.3 | 12.1 | 15.6 |
| TAU2-Telecom | 34.2 | 41.5 | 51.3 |
| IFBench | 39.3 | 54.3 | 65.4 |
| TerminalBench-Hard | 10.1 | 15.2 | 23.9 |
| AA-LCR | 30.4 | 59.9 | 61.2 |
| SciCode | 29.5 | 35.7 | 40.9 |
| AAII Composite Score | 42 (41.53) | 55 (54.79) | 60 (60.31) |
Comparable Models by Configuration
aquif-3.5-Plus (Non-Reasoning) — AAII 42
| Model | AAII Score |
|---|---|
| GPT-5 mini | 42 |
| Claude Haiku 4.5 | 42 |
| Gemini 2.5 Flash Lite 2509 | 42 |
| aquif-3.5-Plus (Non-Reasoning) | 42 |
| DeepSeek V3 0324 | 41 |
| Qwen3 VL 32B Instruct | 41 |
| Qwen3 Coder 480B A35B | 42 |
aquif-3.5-Plus (Reasoning) — AAII 55
| Model | AAII Score |
|---|---|
| GLM-4.6 | 56 |
| Gemini 2.5 Flash 2509 | 54 |
| Claude Haiku 4.5 | 55 |
| aquif-3.5-Plus (Reasoning) | 55 |
| Qwen3 Next 80B A3B | 54 |
aquif-3.5-Max — AAII 60
| Model | AAII Score |
|---|---|
| Gemini 2.5 Pro | 60 |
| Grok 4 Fast | 60 |
| aquif-3.5-Max | 60 |
| MiniMax-M2 | 61 |
| gpt-oss-120B high | 61 |
| GPT-5 mini | 61 |
| DeepSeek-V3.1-Terminus | 58 |
| Claude Opus 4.1 | 59 |
Key Features
Massive Context Windows: Both models support up to 1M tokens, enabling analysis of entire codebases, research papers, and extensive conversation histories without truncation.
Efficient Architecture: Despite offering frontier-level performance, both models maintain exceptional efficiency through optimized mixture-of-experts design and active parameter count of just 3.3B.
Flexible Reasoning (Plus Only): aquif-3.5-Plus provides interchangeable thinking modes—enable reasoning for complex problems, disable for faster inference on straightforward tasks.
Multilingual Support: Native support across English, German, Italian, Portuguese, French, Hindi, Spanish, Thai, Chinese, and Japanese.
Usage Recommendations
aquif-3.5-Plus:
- Complex reasoning requiring flexibility between speed and depth
- Scientific analysis and mathematical problem-solving with thinking enabled
- Rapid-response applications with thinking disabled
- Code generation and review
- Multilingual applications up to 1M token contexts
aquif-3.5-Max:
- Frontier-level problem-solving without compromise
- Advanced research and scientific computing
- Competition mathematics and algorithmic challenges
- Comprehensive code analysis and generation
- Complex multilingual tasks requiring maximum reasoning capability
Setting Thinking Mode (aquif-3.5-Plus)
Toggle between thinking and non-thinking modes by modifying the chat template:
set thinking = true # Enable reasoning mode
set thinking = false # Disable thinking mode (faster inference)
Simply set the variable in your chat template before inference to switch modes. No model reloading required.
Technical Specifications
Both models support:
- BF16 and FP16 precision
- Mixture of Experts architecture optimizations
- Efficient attention mechanisms with optimized KV caching
- Up to 1M token context window
- Multi-head attention with sparse routing
Performance Highlights
aquif-3.5-Plus achieves 82.3% average benchmark performance in thinking mode, surpassing models with 2-4x more total parameters. Non-thinking mode maintains competitive 66.9% performance for latency-sensitive applications.
aquif-3.5-Max reaches 86.2% average performance, matching or exceeding frontier models while maintaining 42.4B total parameters—an extraordinary efficiency breakthrough.
Acknowledgements
- Qwen Team: Base architecture contributions
- Meta Llama Team: Core model foundations
- Hugging Face: Model hosting and training infrastructure
License
This project is released under the Apache 2.0 License. See LICENSE file for details.
Made in 🇧🇷
© 2025 aquif AI. All rights reserved.
🚀 If you find these models useful
Help me test my AI-Powered Quantum Network Monitor Assistant with quantum-ready security checks:
The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : Source Code Quantum Network Monitor. You will also find the code I use to quantize the models if you want to do it yourself GGUFModelBuilder
💬 How to test:
Choose an AI assistant type:
TurboLLM(GPT-4.1-mini)HugLLM(Hugginface Open-source models)TestLLM(Experimental CPU-only)
What I’m Testing
I’m pushing the limits of small open-source models for AI network monitoring, specifically:
- Function calling against live network services
- How small can a model go while still handling:
- Automated Nmap security scans
- Quantum-readiness checks
- Network Monitoring tasks
🟡 TestLLM – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):
- ✅ Zero-configuration setup
- ⏳ 30s load time (slow inference but no API costs) . No token limited as the cost is low.
- 🔧 Help wanted! If you’re into edge-device AI, let’s collaborate!
Other Assistants
🟢 TurboLLM – Uses gpt-4.1-mini :
- **It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
- Create custom cmd processors to run .net code on Quantum Network Monitor Agents
- Real-time network diagnostics and monitoring
- Security Audits
- Penetration testing (Nmap/Metasploit)
🔵 HugLLM – Latest Open-source models:
- 🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.
💡 Example commands you could test:
"Give me info on my websites SSL certificate""Check if my server is using quantum safe encyption for communication""Run a comprehensive security audit on my server"- '"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code on. This is a very flexible and powerful feature. Use with caution!
Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is open source. Feel free to use whatever you find helpful.
If you appreciate the work, please consider buying me a coffee ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
Thank you! 😊
- Downloads last month
- 397