How to choose between open-source and proprietary AI

Open-source and proprietary AI differ in cost, control and capability. This is how to weigh them up.

Open-source AI

Examples: Llama, Mistral, Stable Diffusion, Whisper.

Pros:

• Free to use within the license (some, like Llama, have their own license, so read it)
• Full control over the model
• You can fine-tune it on your own data
• Community support and improvements
• No vendor lock-in

Cons:

• Deploying it takes machine learning skills
• It may need a lot of computing power
• Quality varies widely
• No SLA or guaranteed support

Proprietary AI

Examples: OpenAI's GPT models, Anthropic's Claude, Midjourney.

Pros:

• Often the most capable models available
• Easy to use through an API
• No infrastructure to run
• Regular updates
• Commercial support

Cons:

• Ongoing API costs, usually per token
• No control over changes to the model
• Data privacy concerns
• Vendor lock-in
• Rate limits

Buying an AI product

AI products from smaller sellers are usually one of these:

• A fine-tuned open-source model, such as a Llama model trained for your industry
• An app built on a proprietary API, such as a chatbot that uses OpenAI's models
• A custom system that mixes open-source and proprietary parts

Ask for a demo, test it on your own data and read the license (plus the API provider's terms, for an app built on one) before you commit. With escrow, your payment is held until you confirm it works as described.