What is the difference between open-source and closed AI models?

• Closed models are kept private by the company that built them. You use them through an app or an API, but you cannot see or download the model itself. Many leading chatbots work this way.
• Open models (more precisely, "open-weight" models) release the trained model's weights so anyone can download, run and modify them. Examples include Meta's Llama family, Mistral, China's DeepSeek and Qwen, and several Indian models.

Benefits of open models:

• Startups and researchers can build on them cheaply
• Can be run on your own servers, keeping data private
• Can be adapted for Indian languages and local needs

Benefits of closed models:

• Often the most capable at the cutting edge
• Companies can monitor misuse and update safety protections

Concerns about open models include misuse by bad actors, since safety features can be removed once weights are public. Concerns about closed models include dependence on a few big companies, mostly in the US.

For India, open models are important for building sovereign AI, a goal of the IndiaAI Mission.