Model
Qwen3.6 35B A3B
Intelligence rank
#19/238
71Speed rank
#23/238
52sFastest hardware: NVIDIA RTX 6000 Ada 48GB.
Quick Start
Variations
Intelligence / Disk Size
Verified model-weight disk size. Higher intelligence and lower disk size are better.
Intelligence ÷ Disk Size (GB).
- 6.19iq1m · unsloth/Qwen3.6-35B-A3B-GGUF
- 5.63iq2xxs
- 5.34UD-IQ2_M
- 5.24UD-Q2_K_XL
- 4.98iq3xxs
- 4.66UD-IQ3_S
- 4.36UD-Q3_K_S
- 4.00q3km
- 3.94UD-Q3_K_XL
- 3.85iq4xs
- 3.28UD-Q4_K_S
- 3.13UD-Q4_K_M
- 3.07MXFP4_MOE
- 2.98q4kxl
- 2.97nvfp4 · MTP · k=1 · nvidia/Qwen3.6-35B-A3B-NVFP4
- 2.974bit · unsloth/Qwen3.6-35B-A3B-UD-MLX-4bit
- 2.60awq-4bit · Qwen3 Next Mtp · k=4
- 2.48nvfp4 · unsloth/Qwen3.6-35B-A3B-NVFP4-Fast
- 2.25nvfp4 · unsloth/Qwen3.6-35B-A3B-NVFP4
- 2.19nvfp4 · MTP · k=1 · unsloth/Qwen3.6-35B-A3B-NVFP4-Fast
- 2.04nvfp4 · MTP · k=1 · unsloth/Qwen3.6-35B-A3B-NVFP4
- 1.81fp8 · MTP · k=1
- 0.99Default · MTP · k=1
- 0.98Default · Qwen/Qwen3.6-35B-A3B
Hardware
Cost / task speed
Hardware cost / task speed
Model variations
·5171
Any intelligenceEach coloured line is that variation’s cost / task-speed Pareto frontier. Global is the frontier across every qualifying variation; every point is still one original measured configuration.