Monadd AI
monaddLLM is a quality-preserving, storage-backed inference engine. It runs frontier-scale models — 230 billion to 2.8 trillion parameters — at 8-bit / native precision on ordinary PCs and edge servers, with no quality loss and practical generation speeds.
Models
Models in this organization are published in the .monadd format for monaddLLM:
| Model | Base | Total parameters | Activated parameters | Precision |
|---|---|---|---|---|
| Kimi-K2.6-MONADD | moonshotai/Kimi-K2.6 | 1.03T | 32B | Q8_0 |
| GLM-5.1-MONADD | zai-org/GLM-5.1 | 754B | 40B | Q8_0 |
| DeepSeek-V3.1-Terminus-MONADD | deepseek-ai/DeepSeek-V3.1-Terminus | 685B | 37B | Q8_0 |
| Qwen3-235B-A22B-MONADD | Qwen/Qwen3-235B-A22B | 235B | 22B | Q8_0 |
| MiniMax-M2.5-MONADD | MiniMaxAI/MiniMax-M2.5 | 229B | 10B | Q8_0 |
Each model inherits the license of its base model. The .monadd packaging is by Monadd AI.