What is an open-source AI model?

An open-source AI model is one whose trained model files, known as weights, are released publicly so anyone can download, run, study and often modify it. This is different from closed models, such as the main models behind ChatGPT, Claude and Gemini, which you can only use through the company's app or paid API.

Think of it like the difference between buying a meal at a restaurant and getting the recipe. With a closed model, you get the result but cannot see or change how it is made. With an open model, you can run it on your own hardware and adjust it to your needs.

Popular open model families include Meta's Llama, Mistral from France, DeepSeek and Qwen from China, Google's Gemma and OpenAI's open-weight models. New releases arrive frequently, and the best open models have come surprisingly close to closed models on many tasks.

Why would someone use an open model?

1. Privacy and control. You can run the model on your own servers or computer, so sensitive data never leaves your organisation. This matters for hospitals, banks, government bodies and companies handling client data.
2. Cost at scale. Once you have the hardware, running an open model has no per-message fee. For very high volumes, this can be cheaper.
3. Customisation. You can fine-tune an open model on your own data, for example to learn your company's documents, a regional language or a specialised field.
4. Independence. Your product does not depend on another company's pricing changes or policy decisions.
5. Learning and research. Students and researchers can examine how models behave internally.

There are trade-offs too. The very largest open models need expensive GPUs to run well. You are responsible for setup, updates, security and safety filters. And the most advanced closed models often still lead on the hardest reasoning and coding tasks.

It is also worth knowing that "open" varies. Some models come with fully permissive licences. Others have restrictions, such as limits on commercial use for very large companies. Many release the weights but not the training data or full training code, which is why the term open-weight is often more accurate.

For beginners, tools like Ollama and LM Studio make it easy to try a small open model on a decent laptop, which is a great way to understand how these systems work.