Unverified50% confidenceBenchmarkExact time
Qwen 3.6 35B uses a MoE architecture, and when all expert weights are placed on the CPU with only a small amount of weights kept on the GPU, it can maintain a generation speed of about 40 tokens/second.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/29/2026
First Seen
Valid until: 1/27/2027
Sources
Related Entities
Related Claims
UnverifiedQwen3.5-4B can run efficiently on a single consumer-grade GPU (such as the RTX 4090), and after quantization can even be deployed on edge devices.70% similarUnverifiedOn a desktop CPU of similar performance, the generation speed of a 7B Q4 quantized model is roughly between a few tokens and a dozen-plus tokens per second.70% similarUnverifiedUnder a dual-Spark TP2 configuration, Qwen 3.8 Flash achieves an inference speed of 55–70 tokens/s and supports 4 fully concurrent agents running simultaneously.69% similarUnverifiedThe N-Gram Embedding parameters of Qwen3.8-Flash-Next account for approximately 51B and can be offloaded to CPU memory to reduce GPU VRAM requirements.68% similarUnverifiedA project called Qwenputer embeds the behavior of a 32-bit RISC-V CPU into the weights of the Qwen3.5-4B model, enabling the model to execute RISC-V instructions rather than calling an external emulator.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1112078API
curl https://kongchang.com/api/v1/knowledge/claims/1112078MCP
get_claim(id=1112078)