Open Source AI
Also known as: Open-Source Models, Open Weights, Open AI Models
AI models and systems released with publicly available weights, code, and documentation, allowing inspection, modification, and redistribution.
Open source AI refers to models released with publicly accessible weights and code, enabling anyone to run, study, modify, and share them.
Spectrum of Openness
| Level | What’s Shared |
|---|---|
| Fully open | Weights, code, training data, methodology |
| Open weights | Model weights and inference code |
| API-only | Just an interface, nothing inspectable |
Notable Open Models
- Llama (Meta): Widely adopted foundation model
- Mistral: European open-weight models
- Stable Diffusion: Open image generation
- Whisper (OpenAI): Open speech recognition
Benefits
- Transparency and auditability
- Innovation and research access
- Reduced vendor lock-in
- Local/private deployment options
Concerns
- Misuse potential (no guardrails)
- Compute requirements to run
- Sustainability of development
- “Open washing”—marketing without true openness