Qwen 3.6 35B A3B
A large language model in Alibaba's Qwen series that uses a MoE architecture, with 35B total parameters and 3B activated parameters, aimed at efficient inference and local deployment.
Timeline (last 90 days)
All three tested models are MOE architectures at the 30B parameter level: Gemma 4 has 26B total parameters with 4B active, GLM 4.7 Flash has 30B total parameters with 3B active, and Qwen 3.6 35B A3B has 35B total parameters with 3B active.
Qwen 3.6 35B A3B was basically completed on its first generation; the scene included complete elements such as air conditioning units, traffic lights (which can automatically switch colors), trash cans, streetlights, vending machines, and roads, though the bicycle still had flaws.
The overall performance ranking is: Qwen 3.6 35B A3B first, GLM 4.7 Flash second, and Gemma 4 last.
For the comparison setup Qwen 3.6 35B A3B on llama.cpp, with a VRAM-constrained expert layout and MTP/Vision disabled, prompt processing is 630-680 t/s and generation is 27-35 t/s.
All Facts (4)
All three tested models are MOE architectures at the 30B parameter level: Gemma 4 has 26B total parameters with 4B active, GLM 4.7 Flash has 30B total parameters with 3B active, and Qwen 3.6 35B A3B has 35B total parameters with 3B active.
50%UnverifiedQwen 3.6 35B A3B was basically completed on its first generation; the scene included complete elements such as air conditioning units, traffic lights (which can automatically switch colors), trash cans, streetlights, vending machines, and roads, though the bicycle still had flaws.
50%UnverifiedThe overall performance ranking is: Qwen 3.6 35B A3B first, GLM 4.7 Flash second, and Gemma 4 last.
50%UnverifiedFor the comparison setup Qwen 3.6 35B A3B on llama.cpp, with a VRAM-constrained expert layout and MTP/Vision disabled, prompt processing is 630-680 t/s and generation is 27-35 t/s.
50%