Open-source AI models — led by Meta's Llama 3.2, Mistral's Mixtral, DeepSeek's R2, and Stability AI's Stable Diffusion — now power 38% of enterprise AI applications, up from 15% in 2024.
The shift is driven by cost (no per-token API fees), privacy (models run on private infrastructure), and customisation (open weights allow fine-tuning for specific domains).
The trend threatens the business models of closed-source providers like OpenAI and Anthropic, which charge for API access. If comparable performance can be achieved with free models, the premium for closed-source APIs may erode.
However, the gap between the best open-source and closed-source models remains significant on the most demanding tasks. Anthropic's Claude and OpenAI's GPT-5 still outperform all open-source alternatives on complex reasoning.