A recent analysis explains how open AI models — often called “open-weight models” — are rapidly gaining traction as businesses look for alternatives to proprietary systems like ChatGPT, Gemini, and Claude. While large closed models still dominate public attention, many enterprise leaders are increasingly experimenting with smaller and more customizable open models that offer greater control, flexibility, and cost efficiency. Analysts say companies are realizing that not every AI task requires a massive frontier model running in the cloud.
Open models such as Meta’s Llama, Mistral, DeepSeek, Qwen, and Microsoft Phi are becoming popular because organizations can download, modify, and deploy them on their own infrastructure. This gives businesses stronger governance over sensitive data, lower inference costs, and reduced dependence on external AI providers. Gartner analysts quoted in the report described these models as “blank canvases” that companies can adapt for highly specific internal use cases rather than relying entirely on generalized AI systems.
Another major driver behind the shift is growing concern about vendor lock-in and operational reliability. Recent outages involving large AI providers have pushed CIOs to think more seriously about resilience and fallback systems. Open models can often run locally or on-premise, making them attractive for industries focused on security, privacy, and digital sovereignty. Countries such as France and the UAE are also supporting open-model ecosystems as part of broader sovereign AI strategies designed to reduce dependence on foreign technology platforms.
At the same time, experts caution that open models come with trade-offs. Many still lag behind top proprietary systems in reasoning and general-purpose performance, and they may require significant experimentation and tuning to achieve strong results. Security researchers have also warned that open models can create new risks because vulnerabilities and malicious modifications are harder to control once models are publicly distributed. Still, industry observers increasingly believe the future AI ecosystem will be hybrid — combining frontier proprietary models with specialized open systems optimized for speed, cost, privacy, and enterprise-specific tasks.
