Open-Source AI Models Challenge OpenAI and Google's Market Dominance

In 2024, Meta launched the third version of its model, Llama 3, under a semi-open license allowing commercial use. It quickly topped the list of most downloaded models on the Hugging Face platform, directly challenging OpenAI's paid GPT-4 models.

Chinese company DeepSeek sent shockwaves in January 2025 when it revealed its open model, DeepSeek-R1, which achieved performance competing with GPT-4 at a training cost 96% lower according to its claims, sparking widespread debate about AI markets and the feasibility of massive investment in it.

Companies like France's Mistral and developer communities argue that open-source artificial intelligence narrows the gap between major and small nations in this sector, while opponents warn of the risks of using open models for harmful purposes without adequate restrictions.

This competition reflects a deeper struggle between the "closed AI" model, where major corporations provide limited access for a subscription, and the "open AI" model, which allows developers to run and modify models locally, amid fundamental questions about who owns artificial intelligence and who controls its future.

What do these terms mean?

Open Source: A software philosophy that allows publishing, modifying, and sharing product code freely. It is applied to AI models by publishing model weights and source code so that any developer can access them.

Model Weights: The numbers and mathematical parameters stored inside an AI network that form its memory and knowledge acquired from training; sharing them publicly is what actually makes a model open source.

Hugging Face Platform: An online platform considered the largest repository for open-source AI models, where researchers and companies freely share hundreds of thousands of models and datasets.