Open-Weight
A model whose trained weights are publicly available for download, allowing anyone to run, fine-tune, or build on top of it, distinct from fully open-source (which also includes training code and data).
An open-weight model makes its trained parameters publicly downloadable, usually subject to a license. This is distinct from open-source, which would also include the training code and data. Meta's Llama models, Mistral's models, and Microsoft's Phi series are open-weight: you can download and run them, but you can't replicate the training process.
Open-weight models have transformed the AI landscape by enabling a global ecosystem of fine-tunes, specializations, and applications. The Hugging Face Hub hosts thousands of fine-tuned variants of Llama alone, created by researchers and practitioners worldwide. This ecosystem accelerates progress and democratizes access to powerful AI capabilities.
The licenses attached to open-weight models vary significantly. Llama 4's Community License prohibits competing AI service providers from using it. Mistral Small uses the permissive Apache 2.0 license. Understanding license terms is essential before building commercial applications on open-weight models.
Read next
Open-Weight vs Open-Source Models: What's the Difference?
Why 'open-source AI' is often a misleading term, and what it actually means when a model is open-weight, what's included, what's not, and why it matters for developers.
Open Source vs Closed LLMs: Which Is Right for You?
A practical analysis of open-weight versus proprietary AI models, comparing capability, cost, privacy, control, and real-world tradeoffs for 2025.
Meta AI: How Open-Source Is Reshaping the AI Landscape
Meta's Llama series has become the foundation of the open AI ecosystem, here's the full story of how a social media company became the open-weight AI champion.