Explore AI
Model families
Every family publishes weights. Almost none of them are open source. Here is what each one is actually good at.
Qwen
Alibaba
Mostly Apache-2.0
Best for: general, coding, documents
Llama
Meta
Llama Community Licence — restricted
Best for: general, documents
DeepSeek
DeepSeek
MIT
Best for: coding, agents
Gemma
Google
Apache-2.0
Best for: general, documents, offline
Mistral
Mistral AI
Apache-2.0 on smaller models; research or commercial terms on larger
Best for: general, speech
GLM
Zhipu AI (Z.ai)
Mixed — check each release
Best for: general, agents, documents
Kimi
Moonshot AI
Modified MIT on recent releases
Best for: documents, coding, agents
MiniMax
MiniMax
Mixed, often non-commercial
Best for: images, general
Phi
Microsoft
MIT
Best for: general, offline
OLMo
Allen Institute for AI
Apache-2.0, with training data and code released
Best for: general
Nemotron
NVIDIA
Mixed — often open
Best for: general, agents
Command
Cohere
Non-commercial on published weights
Best for: documents
How to read these
The licence line is the important one. Most families here publish their weights, but "publishing weights" is not the same as "open source", and it is definitely not the same as "free to use in a product". Several carry non-commercial terms or restrictions on scale. Check before you build.
Origin matters for some people. Where a model was trained and by whom is a real consideration for regulated work, and pretending otherwise would be unhelpful.
Every family has a weak spot. None of them is best at everything, which is why this page exists rather than a single recommendation.