FLUX.2 Klein 4B
Apache 2.0Black Forest Labs · 4B · Apache 2.0
Fastest open FLUX.2 — sub-second text-to-image and multi-reference editing on consumer GPUs
2.5GB·32K ctx·
M2 Ultra 能跑的本機 AI 模型
為 M2 Ultra 評級的開源 LLM、程式、図像与影片模型。
GPUM2 UltraVRAMBWRAMCores—
最適合以下装置的開源模型： 你的装置
跑得動写程式的模型吗？
対話、程式与推理
跑得動図像模型吗？
在本機生成図像
跑得動影片模型吗？
在本機生成影片
跑得動超小模型吗？
装置端，4B 以下
这台装置的資源非常有限
目前没有能順跑的模型。切到 全部浏覧 看最軽量的選項。
8 個月前
Wan 2.2 TI2V 5B
Apache 2.0Alibaba · 5B · Apache 2.0
Unified text/image-to-video — the local sweet spot under Apache 2.0
3.1GB·4K ctx·
Z-Image Turbo
Apache 2.0Alibaba · 6B · Apache 2.0
8-step distilled image model — photorealism and bilingual text on 16GB cards
3.6GB·4K ctx·
HunyuanVideo 1.5
Tencent Hunyuan CommunityTencent · 8.3B · Tencent Hunyuan Community
Compact cinematic video model — strong faces and motion on a single 4090
4.8GB·4K ctx·
Qwen3-VL 8B
Apache 2.0Alibaba · 8.8B · Apache 2.0
The community-favourite local VLM — superb OCR, receipts & captioning
5GB·256K ctx·
Qwen 3.5 9B
Apache 2.0Alibaba · 9B · Apache 2.0
Multimodal Qwen 3.5 mid-size
22 AA·5.1GB·32K ctx·
LTX 2.3
LTX-2 CommunityLightricks · 19B · LTX-2 Community
Open 4K video with native stereo audio — text, image and video-to-video
10.2GB·4K ctx·
Qwen Image 2512
Apache 2.0Alibaba · 20B · Apache 2.0
Open text-to-image with strong English and Chinese typography
10.7GB·4K ctx·
GPT-OSS 20B
Apache 2.0OpenAI · 21B · Apache 2.0
OpenAI's open-weight MoE with configurable reasoning
15 AA·11.3GB·128K ctx·
Gemma 4 26B-A4B IT
GemmaGoogle · 27B · Gemma
Gemma 4 MoE instruct model (official)
26 AA·14.3GB·256K ctx·
Qwen 3.8 27B
Apache 2.0Alibaba · 27B · Apache 2.0
Flagship dense Qwen 3.8 — native multimodal all-rounder with video understanding
52 AA·14.3GB·256K ctx·
Wan 2.2 T2V A14B
Apache 2.0Alibaba · 27B · Apache 2.0
Flagship open Wan 2.2 — 14B-active MoE for photoreal text-to-video
14.3GB·4K ctx·
Muse Glimmer 30B
Apache 2.0Meta · 30B · Apache 2.0
Open agentic 30B distilled from Muse Spark — tool use, vision and local recovery on a single GPU
35 AA·15.9GB·128K ctx·
Qwen3-VL 30B-A3B
Apache 2.0Alibaba · 31B · Apache 2.0
Efficient vision MoE — 3B active, strong temporal & document understanding
16.4GB·256K ctx·
FLUX.2 Dev
FLUX Non-CommercialBlack Forest Labs · 32B · FLUX Non-Commercial
Flagship open-weight FLUX.2 — text-to-image and multi-reference editing up to 4MP
16.9GB·32K ctx·
MiniMax H3
MiniMax CommunityMiniMax · 33B · MiniMax Community
Open video generation — text/image to 2K video with native stereo audio
17.4GB·32K ctx·
Agents-A1 35B-A3B
Apache 2.0InternScience · 35B · Apache 2.0
Efficient multimodal agentic MoE for long-horizon search, engineering and scientific research
18.4GB·256K ctx·
Ornith 1.0 35B-A3B
MITDeepReinforce · 35B · MIT
Agentic coding MoE with a 3B active working set and self-improving training
18.4GB·256K ctx·
Qwen 3.6 35B-A3B
Apache 2.0Alibaba · 36B · Apache 2.0
Big-model quality at 3B-active speed — the mid-hardware sweet spot
32 AA·18.9GB·256K ctx·
HunyuanImage 3.0 Instruct
Tencent Hunyuan CommunityTencent · 80B · Tencent Hunyuan Community
Reasoning image model — prompt rewrite, chain-of-thought and image-to-image editing
41.5GB·4K ctx·
Llama 4 Scout 17B
Llama 4 CommunityMeta · 109B · Llama 4 Community
MoE with 16 experts, 17B active params
10 AA·56.3GB·128K ctx·
GPT-OSS 120B
Apache 2.0OpenAI · 117B · Apache 2.0
OpenAI's flagship open-weight MoE — 52.6% SWE-bench
24 AA·60.4GB·128K ctx·
Mistral Small 4 119B
Apache 2.0Mistral AI · 119B · Apache 2.0
Sparse Mistral Small 4 — 6.5B active, strong local all-rounder
20 AA·61.5GB·256K ctx·
Qwen 3 VL 235B-A22B
Apache 2.0Alibaba · 235B · Apache 2.0
Flagship vision-language MoE — frontier multimodal reasoning and agentic GUI control
120.9GB·256K ctx·
Hy3
Apache 2.0Tencent · 295B · Apache 2.0
Production-focused agentic MoE with strong coding, tool use and long-context reasoning
42 AA·151.6GB·256K ctx·
MiniMax M3
MiniMax CommunityMiniMax · 428B · MiniMax Community
Native multimodal MoE — understands text, image and long video with 1M context
45 AA·219.7GB·1024K ctx·
GLM-5.3
MITZ.ai · 753B · MIT
Same 753B / 40B-active base as GLM-5.2 — post-training lifts coding and long-horizon agents, 1M context
386.2GB·1024K ctx·
LongCat 2.0
MITMeituan · 1.6T · MIT
Frontier-scale agentic and coding MoE with sparse attention and native 1M context
34 AA·820.1GB·1024K ctx·
DeepSeek V4 Pro
MITDeepSeek · 1.6T · MIT
Flagship V4 MoE — 49B active, 1M context
820.1GB·1024K ctx·
Qwen 3.8 2.4T-A95B
QwenAlibaba · 2.4T · Qwen
Frontier Qwen 3.8 MoE — 95B active, 1M context
58 AA·1229.8GB·1024K ctx·
Kimi K3
KimiMoonshot AI · 2.8T · Kimi
Frontier 2.8T multimodal MoE — 104B active, native video understanding, 1M context
60 AA·1424.5GB·1024K ctx·
全部模型
Qwen 3 0.6B
Apache 2.0Alibaba · 0.6B · Apache 2.0
Ultra-light Qwen 3 model for constrained devices
0.8GB·32K ctx·
Qwen 3.5 0.8B
Apache 2.0Alibaba · 0.8B · Apache 2.0
Ultra-tiny model for embedded and edge
5* AA·0.9GB·32K ctx·
Llama 3.2 1B
Llama 3.2 CommunityMeta · 1B · Llama 3.2 Community
Meta's smallest Llama for edge devices
1GB·128K ctx·
Gemma 3 1B
GemmaGoogle · 1B · Gemma
Google's tiny Gemma for on-device
1GB·32K ctx·
Wan 2.1 T2V 1.3B
Apache 2.0Alibaba · 1.3B · Apache 2.0
Tiny open text-to-video — 480p clips on 8GB consumer GPUs
1.2GB·4K ctx·
Qwen 2.5 Coder 1.5B
Apache 2.0Alibaba · 1.5B · Apache 2.0
Ultra-lightweight coding model
1.3GB·32K ctx·
DeepSeek R1 1.5B
MITDeepSeek · 1.5B · MIT
Tiny reasoning model distilled from R1
1.3GB·64K ctx·
Qwen 3 1.7B
Apache 2.0Alibaba · 1.7B · Apache 2.0
Compact multilingual Qwen 3
1.4GB·32K ctx·
Qwen 3.5 2B
Apache 2.0Alibaba · 2B · Apache 2.0
Small multimodal Qwen 3.5
7* AA·1.5GB·32K ctx·
Llama 3.2 3B
Llama 3.2 CommunityMeta · 3B · Llama 3.2 Community
Lightweight Llama for mobile and edge
2GB·128K ctx·
SmolLM3 3B
Apache 2.0HuggingFace · 3B · Apache 2.0
Lightweight multilingual reasoning
2GB·128K ctx·
Granite 4.1 3B
Apache 2.0IBM · 3B · Apache 2.0
Compact enterprise model for edge and constrained environments
2GB·128K ctx·
Ministral 3 3B
Apache 2.0Mistral AI · 3B · Apache 2.0
Current-gen tiny Ministral — edge chat with 256K context
7 AA·2GB·256K ctx·
Phi-4 Mini Reasoning
MITMicrosoft · 3.8B · MIT
Lightweight reasoning model
2.4GB·16K ctx·
Gemma 3 4B
GemmaGoogle · 4B · Gemma
Multimodal Gemma with 128K context
2.5GB·128K ctx·
Qwen 3.5 4B
Apache 2.0Alibaba · 4B · Apache 2.0
Small multimodal Qwen 3.5
20* AA·2.5GB·32K ctx·
Qwen3-VL 4B
Apache 2.0Alibaba · 4.4B · Apache 2.0
Compact dedicated vision-language model — OCR & image chat on edge
2.8GB·256K ctx·
Gemma 4 E2B IT
GemmaGoogle · 5B · Gemma
Gemma 4 efficient instruct model (official)
10* AA·3.1GB·256K ctx·
Qwen 2.5 Coder 7B
Apache 2.0Alibaba · 7B · Apache 2.0
Dedicated coding model
4.1GB·128K ctx·
DeepSeek R1 Distill 7B
MITDeepSeek · 7B · MIT
R1 reasoning distilled into Qwen 7B
4.1GB·64K ctx·
Gemma 4 E4B IT
GemmaGoogle · 8B · Gemma
Gemma 4 balanced instruct model (official)
12* AA·4.6GB·256K ctx·
Llama 3.1 8B
Llama 3.1 CommunityMeta · 8B · Llama 3.1 Community
Meta's versatile 8B — great quality/speed ratio
4.6GB·128K ctx·
Qwen 3 8B
Apache 2.0Alibaba · 8B · Apache 2.0
Qwen 3 with thinking mode support
4.6GB·128K ctx·
Granite 4.1 8B
Apache 2.0IBM · 8B · Apache 2.0
Balanced general-purpose enterprise model
4.6GB·128K ctx·
Ministral 8B
MRLMistral AI · 8B · MRL
Mistral's efficient 8B model
4.6GB·32K ctx·
GLM-4 9B
GLM-4Zhipu AI · 9B · GLM-4
Multilingual model supporting 26 languages with 128K context
5.1GB·128K ctx·
Nemotron Nano 9B v2
NVIDIA OpenNVIDIA · 9B · NVIDIA Open
Hybrid Mamba2 architecture for reasoning
9* AA·5.1GB·128K ctx·
Ornith 1.0 9B
MITDeepReinforce · 9B · MIT
Self-improving agentic coding model optimized for terminal and software engineering tasks
5.1GB·256K ctx·
FLUX.2 Klein 9B
FLUX Non-CommercialBlack Forest Labs · 9B · FLUX Non-Commercial
Higher-quality distilled FLUX.2 — sub-second generation and multi-reference editing
5.1GB·32K ctx·
Gemma 3 12B
GemmaGoogle · 12B · Gemma
Multimodal Gemma with 128K context
6.6GB·128K ctx·
Mistral Nemo 12B
Apache 2.0Mistral AI · 12B · Apache 2.0
Multilingual 12B with 128K context
6.6GB·128K ctx·
Gemma 4 12B IT
Apache 2.0Google · 12B · Apache 2.0
Gemma 4 mid-size instruct — multimodal any-to-any
22* AA·6.6GB·256K ctx·
Phi-4 14B
MITMicrosoft · 14B · MIT
Microsoft's reasoning-focused model
5* AA·7.7GB·16K ctx·
Qwen 3 14B
Apache 2.0Alibaba · 14B · Apache 2.0
Strong all-rounder with thinking mode
7.7GB·128K ctx·
DeepSeek R1 Distill 14B
MITDeepSeek · 14B · MIT
R1 reasoning distilled into Qwen 14B
7.7GB·64K ctx·
Ministral 3 14B
Apache 2.0Mistral AI · 14B · Apache 2.0
Current-gen Ministral mid-size — local assistant with 256K context
11 AA·7.7GB·256K ctx·
LFM2 24B
Liquid AILiquid AI · 24B · Liquid AI
Hybrid MoE with convolution+attention layers — 2.3B active
5* AA·12.8GB·32K ctx·
Devstral Small 2 24B
Apache 2.0Mistral AI · 24B · Apache 2.0
Coding-focused model with 256K context — 68% SWE-bench
12.8GB·256K ctx·
Mistral Small 3.1 24B
Apache 2.0Mistral AI · 24B · Apache 2.0
Multimodal Mistral with vision support
12.8GB·128K ctx·
DiffusionGemma 26B-A4B IT
Apache 2.0Google · 26B · Apache 2.0
Discrete diffusion MoE — 1100+ tok/s on H100, multimodal (text/image/video)
13* AA·13.8GB·256K ctx·
Qwen 3.5 27B
Apache 2.0Alibaba · 27.8B · Apache 2.0
Flagship native multimodal Qwen 3.5
14.7GB·256K ctx·
Qwen 3.6 27B
Apache 2.0Alibaba · 27.8B · Apache 2.0
Flagship dense Qwen 3.6 — native multimodal all-rounder
38 AA·14.7GB·256K ctx·
Qwen 3 30B-A3B
Apache 2.0Alibaba · 30B · Apache 2.0
MoE with only 3.3B active — extremely efficient
15.9GB·128K ctx·
Nemotron 3 Nano 30B
NVIDIA OpenNVIDIA · 30B · NVIDIA Open
MoE with 1M context and 3B active
15 AA·15.9GB·1024K ctx·
Granite 4.1 30B
Apache 2.0IBM · 30B · Apache 2.0
High-capacity enterprise model for complex reasoning and tool use
15.9GB·128K ctx·
North Mini Code
Apache 2.0Cohere · 30B · Apache 2.0
Open agentic coding MoE with 3B active — built for software engineering and terminal tasks
15.9GB·256K ctx·
Qwen 3 Coder 30B-A3B
Apache 2.0Alibaba · 30B · Apache 2.0
Efficient agentic coding MoE — 3B active, 256K context
15.9GB·256K ctx·
Qwen 3 32B
Apache 2.0Alibaba · 32B · Apache 2.0
Qwen 3 flagship dense model
16.9GB·128K ctx·
DeepSeek R1 Distill 32B
MITDeepSeek · 32B · MIT
R1 reasoning distilled into Qwen 32B — sweet spot
16.9GB·64K ctx·
OLMo 2 32B
Apache 2.0Allen AI · 32B · Apache 2.0
Fully open research model by Allen AI
16.9GB·4K ctx·
Gemma 4 31B IT
GemmaGoogle · 33B · Gemma
Gemma 4 flagship instruct model (official)
30 AA·17.4GB·256K ctx·
Gemma 4 31B
GemmaGoogle · 33B · Gemma
Gemma 4 flagship base model (official)
17.4GB·256K ctx·
Command R 35B
CC BY-NC 4.0Cohere · 35B · CC BY-NC 4.0
Optimized for retrieval-augmented generation
18.4GB·128K ctx·
Qwen 3.5 35B-A3B
Apache 2.0Alibaba · 35B · Apache 2.0
Efficient multimodal MoE with 3B active
18.4GB·256K ctx·
Mixtral 8x7B
Apache 2.0Mistral AI · 47B · Apache 2.0
MoE with 12.9B active params
24.6GB·32K ctx·
Llama 3.3 70B
Llama 3.3 CommunityMeta · 70B · Llama 3.3 Community
Best open model at 70B class
9* AA·36.4GB·128K ctx·
Qwen 3 Next 80B-A3B
Apache 2.0Alibaba · 80B · Apache 2.0
High-sparsity MoE — extreme low activation ratio for fast inference at 80B scale
41.5GB·256K ctx·
Qwen 3 Coder Next 80B-A3B
Apache 2.0Alibaba · 80B · Apache 2.0
Ultra-efficient agentic coding MoE optimized for tool-calling coding agents
41.5GB·256K ctx·
HunyuanImage 3.0
Tencent Hunyuan CommunityTencent · 80B · Tencent Hunyuan Community
Largest open image MoE — 13B active, strong long-prompt generation
41.5GB·4K ctx·
GLM-4.5 Air
MITZ.ai · 106B · MIT
Consumer-friendly GLM MoE — 12B active, strong agentic & tool use
54.8GB·128K ctx·
Qwen 3.5 122B-A10B
Apache 2.0Alibaba · 122B · Apache 2.0
Large multimodal MoE
33 AA·63GB·256K ctx·
DeepSeek V4 Flash
MITDeepSeek · 158B · MIT
Efficient long-context V4 — 13B active, 1M context
81.4GB·1024K ctx·
Qwen 3 235B-A22B
Apache 2.0Alibaba · 235B · Apache 2.0
Massive MoE with 22B active — frontier quality
120.9GB·128K ctx·
GLM-4.6
MITZ.ai · 357B · MIT
Large GLM MoE with strong coding and 200K context
183.4GB·195K ctx·
Qwen 3.5 397B-A17B
Apache 2.0Alibaba · 397B · Apache 2.0
Largest multimodal Qwen 3.5 MoE
34 AA·203.9GB·256K ctx·
Llama 4 Maverick 17B-128E
Llama 4 CommunityMeta · 400B · Llama 4 Community
Multimodal MoE with 128 experts — 17B active, 1M context
14 AA·205.4GB·1024K ctx·
Qwen 3 Coder 480B
Apache 2.0Alibaba · 480B · Apache 2.0
Largest open coding MoE — 35B active
246.4GB·256K ctx·
DeepSeek R1
MITDeepSeek · 671B · MIT
Massive MoE reasoning model — 37B active
344.2GB·64K ctx·
DeepSeek V3.2
MITDeepSeek · 685B · MIT
State-of-the-art MoE — 37B active params
351.4GB·128K ctx·
GLM-5
MITZhipu AI · 744B · MIT
MoE with 256 experts, 40B active — frontier-class agentic coding
381.6GB·128K ctx·
GLM-5.2
MITZ.ai · 753B · MIT
Frontier open-weight coder — top SWE-bench, 1M context
53 AA·386.2GB·1024K ctx·
GLM-5.1
MITZhipu AI · 754B · MIT
Improved agentic coding — SOTA SWE-bench Pro, long-horizon tasks
386.7GB·128K ctx·
Kimi K2.6
KimiMoonshot AI · 1.06T · Kimi
Natively multimodal 1T MoE — 32B active, frontier agentic
542.4GB·256K ctx·
找不到模型
換個関鍵字或筛選条件試試