Qwen3.5 - 长风

Qwen3.5 是阿里云在 2026 年初发布的最新一代多模态混合架构大模型。

模型

名称	量化类型	激活参数量	大小	权重	平台
Qwen/Qwen3.5-397B-A17B-FP8	FP8	397B	406G		HuggingFace
Qwen/Qwen3.5-397B-A17B-GPTQ-Int4	GPT-Int4	397B	236G		HuggingFace
Qwen/Qwen3.5-122B-A10B	FP32	122B	250G		HuggingFace
Qwen/Qwen3.5-122B-A10B-FP8	FP8	122B	127G		HuggingFace
Qwen/Qwen3.5-122B-A10B-GPTQ-Int4	GPTQ-Int4	122B	78.9G	68.14Gib	HuggingFace
Qwen/Qwen3.5-35B-A3B	FP32	35B	71.9G		HuggingFace
Qwen/Qwen3.5-35B-A3B-FP8	FP8	35B	37.5G		HuggingFace
Qwen/Qwen3.5-35B-A3B-GPTQ-Int4	GPTQ-Int4	35B	24.5G	21.05Gib	HuggingFace
Qwen/Qwen3.5-27B	FP32	27B	55.6G		HuggingFace
Qwen/Qwen3.5-27B-FP8	FP8	27B	30.9G		HuggingFace
Qwen/Qwen3.5-27B-GPTQ-Int4	GPTQ-Int4	27B	30.3G	27.42Gib	HuggingFace
Qwen/Qwen3.5-9B	FP32	9B	19.3G		HuggingFace
qwen3.5:35b		35B	24GB		Ollama
qwen3.5:122b		122B	81G		Ollama

权重是通过 [[vLLM]] 启动信息中 “Model loading took * GiB memory” 字段而来。