Analyze and compare open-weight AI models
Parameters, experts, memory and the hardware to run it, read straight from the model's files.
- Analyze a model Where its parameters live, layer by layer: experts, attention, precision, context memory.
- Compare two models Side by side and to scale: size, active parameters, layers and hardware needs.
- What fits my GPU? Pick a graphics card or Mac: every model that runs, the file to get, and the context left.
Most downloaded models this month
Hugging Face downloads, past 30 days- Qwen/Qwen3-0.6BDense752M · 596M activeRuns on 8 GB
- Qwen/Qwen3-VL-8B-InstructDense + vision8.77B · 7.57B activeRuns on 8 GB as GGUF
- openai-community/gpt2Dense137MRuns on 8 GB
- Qwen/Qwen3-8BDense8.19BRuns on 8 GB as GGUF
- unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUFMoE30.5B · 3.04B activeRuns on 12 GB
- google/gemma-4-26B-A4B-itMoE + vision25.8BRuns on 24 GB as GGUF
- Qwen/Qwen3.6-35B-A3B-FP8MoE + vision36.0B · 3.01B active24 GB if quantized
- Qwen/Qwen2.5-7B-InstructDense7.62BRuns on 24 GB
- Qwen/Qwen3.5-9BDense + vision9.65B · 8.18B activeRuns on 8 GB as GGUF
- google/gemma-4-31B-itDense + vision31.3B24 GB if quantized
- Qwen/Qwen2.5-0.5B-InstructDense494MRuns on 8 GB
- Qwen/Qwen3.8-27BDense + vision27.8BRuns on 24 GB as GGUF
- Qwen/Qwen2.5-1.5B-InstructDense1.54BRuns on 8 GB
- Qwen/Qwen3-4BDense4.02BRuns on 8 GB as GGUF
- Qwen/Qwen3.8-27B-FP8Dense + vision27.8B24 GB if quantized
- Qwen/Qwen3.5-4BDense + vision4.66BRuns on 8 GB as GGUF
- Qwen/Qwen2.5-VL-7B-InstructDense + vision8.29B · 7.07B activeRuns on 24 GB
- openai/gpt-oss-20bMoE20.9B · 3.61B activeRuns on 16 GB
- zai-org/GLM-5.3-FlashMoE + vision321B · 16.7B activeNeeds data-center GPUs
- ornith-ai/Ornith-1.5-9B-GGUFDense9.20B · 7.94B activeRuns on 8 GB
- Qwen/Qwen3.6-27B-FP8Dense + vision27.8B24 GB if quantized
- openai/gpt-oss-120bMoE117B · 5.13B activeNeeds data-center GPUs
- Qwen/Qwen2.5-3B-InstructDense3.09BRuns on 8 GB
- Qwen/Qwen3-32BDense32.8BRuns on 24 GB as GGUF
- dphn/dolphin-2.9.1-yi-1.5-34bDense34.4B24 GB if quantized
- Qwen/Qwen3.5-2BDense + vision2.27B · 1.94B activeRuns on 8 GB
- lmstudio-community/Qwen3.8-27B-MLX-4bitDense + vision27.4BRuns on 24 GB
- ornith-ai/Ornith-1.5-35B-A3B-GGUFMoE35.5B · 2.95B activeRuns on 32 GB
- lmstudio-community/Qwen3.8-27B-MLX-8bitDense + vision27.4B24 GB if quantized
- lmstudio-community/Qwen3.8-27B-MLX-6bitDense + vision27.4BRuns on 32 GB
- lmstudio-community/Qwen3.8-27B-MLX-5bitDense + vision27.4BRuns on 24 GB
- Qwen/Qwen-72BDense72.3BNeeds data-center GPUs
- Qwen/Qwen3-4B-Instruct-2507Dense4.02BRuns on 12 GB
- Qwen/Qwen3-1.7BDense2.03B · 1.72B activeRuns on 8 GB
- Qwen/Qwen3-VL-4B-InstructDense + vision4.44BRuns on 8 GB as GGUF
- Qwen/Qwen3.6-27BDense + vision27.8BRuns on 24 GB as GGUF
- RadixArk/Kimi-K3-DSpark2.25BRuns on 8 GB
- Qwen/Qwen2.5-7B-Instruct-AWQDense7.62BRuns on 8 GB
- nvidia/NVIDIA-Nemotron-3-Nano-4B-BF163.97BRuns on 12 GB
- Qwen/Qwen3.6-35B-A3BMoE + vision36.0B · 3.01B activeRuns on 24 GB as GGUF
- ornith-ai/Ornith-1.0-9B-GGUFDense8.95B · 7.94B activeRuns on 8 GB
- prism-ml/Ternary-Bonsai-2-27B-ggufDense26.9BRuns on 8 GB
- Qwen/Qwen3-VL-2B-InstructDense + vision2.13B · 1.72B activeRuns on 8 GB
- datalab-to/chandra-ocr-2Dense + vision5.30B · 4.33B activeRuns on 12 GB
- Qwen/Qwen2.5-Coder-7B-InstructDense7.62BRuns on 8 GB as GGUF
- Qwen/Qwen3-14B-AWQDense14.8BRuns on 12 GB
- Qwen/Qwen3-VL-8B-Instruct-FP8Dense + vision8.77B · 7.57B activeRuns on 16 GB
- Qwen/Qwen2.5-VL-3B-InstructDense + vision3.75B · 3.09B activeRuns on 12 GB
- deepseek-ai/DeepSeek-V3.2MoE685B · 37.5B activeNeeds data-center GPUs
- Qwen/Qwen3.5-0.8BDense + vision873M · 773M activeRuns on 8 GB
- Qwen/Qwen2.5-14B-InstructDense14.8B12 GB if quantized
- deepseek-ai/DeepSeek-OCRMoE + vision3.34B · 672M activeRuns on 8 GB
- JonathanColetti/Qwen3.8-27B-Uncensored-GGUFDense27.3BRuns on 12 GB
- HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUFDense27.3BRuns on 12 GB
- ggml-org/Qwen3.8-27B-GGUFDense26.9BRuns on 24 GB
- Qwen/Qwen2.5-32B-InstructDense32.8BRuns on 24 GB as GGUF
- farbodtavakkoli/OTel-2.0-LLM-31B-ITDense + vision31.3B24 GB if quantized
- ornith-ai/Ornith-1.0-35B-GGUFMoE34.7B · 2.95B activeRuns on 24 GB
- moonshotai/Kimi-K3MoE + vision2.78T · 104B activeNeeds data-center GPUs
- huihui-ai/Huihui-Qwen3.8-27B-abliterated-GGUFDense27.3BRuns on 8 GB
- HuggingFaceTB/SmolLM2-135MDense135MRuns on 8 GB
- distilbert/distilgpt2Dense88.2MRuns on 8 GB
- Qwen/Qwen2.5-14B-Instruct-AWQDense14.8BRuns on 12 GB
- Qwen/Qwen2-VL-2B-InstructDense + vision2.21B · 1.54B activeRuns on 8 GB
- vikhyatk/moondream2Dense + vision1.93B · 1.37B activeRuns on 8 GB
- ornith-ai/Ornith-1.0-35BMoE + vision35.1B · 2.95B activeRuns on 24 GB as GGUF
- Qwen/Qwen2.5-VL-7B-Instruct-AWQDense + vision8.29B · 7.07B activeRuns on 8 GB
- antirez/deepseek-v4-ggufMoE284B · 13.0B activeNeeds data-center GPUs
- Qwen/Qwen3.5-35B-A3BMoE + vision36.0B · 3.01B activeRuns on 24 GB as GGUF
- HauhauCS/Gemma-4-E4B-Uncensored-HauhauCS-Aggressive7.52BRuns on 8 GB
- DavidAU/Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP-GGUFDense9.20B · 7.94B activeRuns on 8 GB
- CMSManhattan/JiRackUltra_1bDense1.78B · 1.54B activeRuns on 8 GB
- Qwen/Qwen3.5-27BDense + vision27.8B24 GB if quantized
- Qwen/Qwen3-14BDense14.8BRuns on 12 GB as GGUF
- zai-org/GLM-4.7-FlashMoE31.2B · 3.58B active24 GB if quantized
- Qwen/Qwen2.5-Coder-14B-InstructDense14.8B12 GB if quantized
- DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER-MAX-MTP-GGUFDense27.3BRuns on 16 GB
- Qwen/Qwen3-30B-A3BMoE30.5B · 3.04B activeRuns on 24 GB as GGUF
- Qwen/Qwen3-8B-AWQDense8.19BRuns on 8 GB
- mistralai/Mistral-7B-Instruct-v0.2Dense7.24BRuns on 24 GB
- zai-org/GLM-5.3MoE753B · 40.3B activeNeeds data-center GPUs
- 0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUFDense26.9BRuns on 12 GB
- llava-hf/llava-1.5-7b-hfDense + vision7.06BRuns on 24 GB
- unsloth/Qwen3.6-27B-NVFP4Dense + vision27.8BRuns on 32 GB
- Qwen/Qwen2-VL-7B-Instruct-AWQDense + vision8.29B · 7.07B activeRuns on 8 GB
- Qwen/Qwen3.5-35B-A3B-FP8MoE + vision36.0B · 3.01B active24 GB if quantized
- Qwen/Qwen2.5-0.5BDense494MRuns on 8 GB
- autotrust/JEV-27B-VLDense + vision27.8B24 GB if quantized
- Qwen/Qwen3-4B-BaseDense4.02BRuns on 12 GB
- unsloth/Qwen3.5-9B-GGUFDense8.95B · 7.94B activeRuns on 8 GB
- Qwen/Qwen3-1.7B-BaseDense1.72BRuns on 8 GB
- HuggingFaceTB/SmolLM2-135M-InstructDense135MRuns on 8 GB
- Qwen/Qwen2.5-Coder-32B-Instruct-AWQDense32.8BRuns on 24 GB
- MiniMaxAI/MiniMax-M2.7MoE229B · 10.4B activeNeeds data-center GPUs
- TinyLlama/TinyLlama-1.1B-Chat-v1.0Dense1.10BRuns on 8 GB
- ornith-ai/Ornith-1.5-397B-GGUFMoE403B · 16.3B activeNeeds data-center GPUs
- Qwen/Qwen3-Coder-Next-FP8MoE79.7B · 3.56B activeNeeds data-center GPUs
- unsloth/Inkling-Small-GGUFMoE264B · 11.2B activeNeeds data-center GPUs
- peculiar-ragdoll/Dirk-Qwen3.8-27B-GGUFDense27.3BRuns on 12 GB
- apple/OpenELM-1_1B-Instruct1.08BRuns on 8 GB
Trending this month
Past 30 days
Trending today
Past 24 hours
New this week
Most downloaded of this week's releases- tencent/WeVisOne-VL-4BDense + vision4.83B · 4.02B activeRuns on 12 GB
- KRAFTON/Raon-1-mini-previewMoE28.8B · 3.65B activeRuns on 24 GB as GGUF
- tencent/WeVisOne-VL-2BDense + vision2.44B · 1.72B activeRuns on 8 GB
- ATH-MaaS/OvisOCR3Dense + vision853M · 752M activeRuns on 8 GB
- baidu/Unlimited-OCR-2MoE + vision3.34B · 680M activeRuns on 8 GB
- gugujiji/novel-humanize-4bDense4.02BRuns on 12 GB
- XHToken/Spark-X2.6-4B-FP84.11BRuns on 8 GB
- HuggingFaceBio/Carbon-XDense14.8B12 GB if quantized
Newest Guides
All guides- The best local LLMs in 2026, by what fits your GPUThe best local LLMs for every graphics card from 8 GB to 32 GB, measured from the weight files: which models fit, which file to download, and how much quality each tier keeps.
- The most downloaded AI models that fit on your GPUThe AI models people actually download, and which of them fit an RTX 5090, 4090, 3090 or 5080 — with the exact file each card can run and how much context is left over.
- The largest AI models that fit on your GPUHow big a model an RTX 5090, 4090, 3090 or 5080 can actually hold — and why the answer is the same 35B model at four different levels of compression.