本地大模型排行榜

本地大模型排行榜 – 能在家跑的开放模型

每周从 Hugging Face 热门模型更新，按独立的智力分数排名，可按显存筛选。

截至 2026-W40，本地模型榜首是 DeepSeek R1 Distill Qwen 32B GGUF（智力分数 8.4）；下载最多的是 Qwen3 VL 8B Instruct。数据每周从 Hugging Face 与 Artificial Analysis 更新。

#	模型	参数量	智力分数	Coding	30 天下载
1	DeepSeek R1 Distill Qwen 32B GGUF	32.8B	8.4	—	0.05M
2	Qwen3 VL 8B Instruct	8.8B	7.3	—	17.08M
3	Llama 3.1 8B	8.0B	6.9	5.4	0.58M
4	Qwen2.5 Coder 32B Instruct	32.8B	6.7	—	1.05M
5	Qwen2.5 Coder 32B Instruct abliterated GGUF	32.8B	6.7	—	0.01M
6	Qwen2.5 Coder 7B Instruct GGUF	7.6B	5.8	—	0.30M
7	Qwen2.5 Coder 7B Instruct abliterated GGUF	7.6B	5.8	—	0.04M
8	Llama 2 7b hf	6.7B	5.7	—	0.79M
9	Qwen3 8B	8.2B	—	—	11.69M
10	Qwen2.5 7B Instruct	7.6B	—	—	9.64M
11	Qwen3.8 27B	27.8B	—	—	7.02M
12	Mistral 7B Instruct v0.3	7.2B	—	—	2.18M
13	Qwen3 14B	14.8B	—	—	2.16M
14	Qwen3.8 27B TURBO Fable Cold Fusion 735 882 Heretic Uncensored NEO CODER MAX MTP GGUF	26.9B	—	—	1.73M
15	Qwen2.5 Coder 14B Instruct	14.8B	—	—	1.69M
16	Meta Llama 3 8B Instruct	8.0B	—	—	1.05M
17	Apertus v1.5 8B	8.9B	—	—	0.49M
18	Meta Llama 3 8B	8.0B	—	—	0.23M
19	Swift 1.5 Qwen3.8 27B GSQ RCO GGUF	26.9B	—	—	0.12M
20	Qwen2.5 Coder 14B Instruct abliterated GGUF	14.8B	—	—	0.04M
21	Qwen3 VL 32B Instruct ultra uncensored heretic H3 GGUF	25.2B	—	—	0.04M
22	MiniMax H3 TextEncoder Qwen3VL 32B abliterated GGUF	26.5B	—	—	0.03M
23	K2 Horizon 7B	9.0B	—	—	0.02M
24	Qwen3.8 Flash Next MLX Serve mixed 4 8bit	133.2B	—	—	0.01M
25	L3.2 Rogue Creative Instruct Uncensored Abliterated 7B GGUF	7.5B	—	—	0.01M
26	OrcaSAQ 2 Cyber 27B Uncensored GGUF	27.3B	—	—	0.01M
27	DeepHat V1 7B	7.6B	—	—	0.01M
28	CLM v0.1 8B	—	—	—	0.00M
29	Xyntetik Kvist 14B	14.4B	—	—	0.00M
30	llm jp 4.1 32b a3b thinking gguf	32.1B	—	—	0.00M
31	llm jp 4.1 32b a3b thinking	32.1B	—	—	0.00M
32	Wan2.2 Animate 2 14B	—	—	—	—
33	Qwen3 VL 32B Heretic MiniMax H3 NVFP4	—	—	—	—
34	Qwen3 VL 32B Ultra Heretic H3 ComfyUI INT8 ConvRot	—	—	—	—
35	Qwen3 VL 32B Heretic MiniMax H3 NVFP4	—	—	—	—