Qwen 3.6 35B-A3B
Overview
- 35B-parameter mixture-of-experts language model from the Qwen Series
- A3B routing variant activates ~3B parameters per token, maximizing throughput vs. memory tradeoffs
- Architecture: Sparse MoE with dense attention, optimized expert gating, and instruction-tuned reasoning/code capabilities
- Training: Multilingual corpus, heavy code synthesis, aligned for complex tool-use and long-context retention
Specialized Variants: Qwopus Coder
- Qwopus 3.6-35B-A3B-Coder: A specialized derivative developed by Jackrong, built on the Qwen 3.6-35B A3B base
- Agentic Self-Correction: Features “thinking-off” capabilities and token-efficient coding agent behaviors
- MTP Integration: Utilizes MTP (Multi-Token Prediction) to drive efficiency and enable the model to fix its own bugs
- Performance: Demonstrated high throughput (up to 160 s) in coding tasks while maintaining low VRAM footprint
- See Qwopus Coder: Agentic Code Self-Correction and MTP-Driven Efficiency for detailed analysis
Local Deployment & Performance
- Validated inference on 6GB vram constraints via llamacpp GGUF pipelines
- Achieving Fast 35B MoE AI Model Performance on 6GB VRAM with Llama.cpp documents:
- Successful execution on 8-year-old consumer GPU hardware
- Q4_K_M / Q