Major AI model families whose weights are published openly on Hugging Face. Each link goes to the publisher's official page. Licenses vary, and some are restrictive or require you to accept terms, so read the model card before you use one.
Qwen3 and newer models from the Qwen team, from 0.6B up. Qwen3-14B is Apache-2.0 (code: github.com/QwenLM/Qwen3).
DeepSeek-V3.2, R1 and V4.1-Flash. V3.2 and R1 are MIT licensed per their model cards.
Open-weight reasoning models. gpt-oss-20b is Apache-2.0 (code: github.com/openai/gpt-oss).
The Gemma 4 family, including 31B and 26B-A4B instruct models. gemma-4-31B-it is listed as Apache-2.0.
The Llama 3.x family under the Llama community license. Downloads are gated behind accepting the terms.
Kimi K2.x and K3 models (custom license, see the model cards; code: github.com/MoonshotAI/Kimi-K2).
GLM-5.3 and GLM-5.3-Flash (see the model cards for license terms).
MiniMax-M2.x and H3 models (see the model cards for license terms).
Mistral 7B and newer open models. Mistral-7B-Instruct-v0.3 is Apache-2.0.
NVIDIA Nemotron 3 models, from Nano 30B-A3B up to Ultra.
A small language model. phi-4 is MIT licensed.
Fully open models from Ai2, including Olmo-3-7B-Instruct (code: github.com/allenai/OLMo).
IBM Granite 4.x models, with GGUF builds available.
Falcon and Falcon3 models from TII (see the model cards for license terms).
Tiny SmolLM2 and SmolVLM models from Hugging Face. SmolLM2-135M-Instruct is Apache-2.0.
Runs open-weight models locally.
The C/C++ inference engine for GGUF models.
Licenses are as shown on each Hugging Face model card on Nov 22, 2026. This site hosts no model weights and links only to the publishers' official pages.