Open-source (more precisely, open-weight) AI models are ones you can download, run and often modify yourself — like Meta’s Llama, Mistral’s models and DeepSeek. They enable privacy, control and lower cost.

What is open-source AI?

Open-weight models have their trained parameters released publicly, so anyone can download and run them on their own hardware or cloud. Examples include Meta’s Llama, Mistral’s models, and DeepSeek. Note that ‘open’ varies — some licences restrict commercial use or large-scale deployment, so check each model’s terms.

Why it matters

Open models let you run AI privately (data stays on your infrastructure), avoid per-use fees, customise and fine-tune for your needs, and avoid lock-in to a single vendor. This is attractive for businesses with privacy, cost or sovereignty concerns, and for developers who want full control.

The trade-offs

Open models can lag the very best closed models on some tasks, require technical setup and computing resources to run well, and you take on responsibility for safety and updates. For most consumers, hosted assistants are easier; for organisations needing control and privacy, open-weight models (self-hosted or via platforms that support bring-your-own-model) are a powerful option.

Once you are running several AI tools at once, consolidation becomes a real question. Platforms such as osFoundry — an agentic AI platform that is model-agnostic and can self-host — exist to bring chat, agents, knowledge and apps under one roof.

This article is general information, not professional, legal or financial advice. AI tools, prices and availability change fast — verify current details on the official source before you rely on them.