What are ‘open’ AI models?
The tech industry faced something of a reckoning in late 2022 with the arrival of OpenAI’s ChatGPT , a simple chatbot that could answer questions, seemingly by magic. The early frontier large language models (LLMs) from OpenAI and rivals like Anthropic and Google could reason, solve problems, and answer detailed queries put to them using natural language. But no one knew exactly what was in the…
Open AI models, such as ChatGPT, gained attention in late 2022 due to their ability to reason, solve problems, and answer detailed queries using natural language. However, the inner workings of these large language models (LLMs) remained a mystery. In response, AI models like Meta's Llama and Mistral were created, allowing them to be freely downloaded and modified for specific applications.
DeepSeek's open model V3, trained at a lower cost than US frontier labs, and reasoning model R1, released in late 2024 and January 2025 respectively, were also part of this trend.
Open models are gaining ground among enterprises for reasoning, agentic, and physical AI. Enterprises can customize open-weight models to their specific needs, audit them, and deploy them in-house. These models help provide assurance and control over data, evaluation, and adaptation. Open-weight models hide certain information, such as code and training data, so they may not be fully open-source. Nonetheless, companies like Alibaba and Moonshot's Kimi are adopting open models for their specific applications.
Open models offer enterprises greater control, security, and customization options. They can be run internally, avoiding cloud vulnerabilities and providing additional security for regulated industries. Enterprises can inspect weights, audit training data, and air-gap deployments, ensuring better governance and control over AI systems. However, open models may eventually become less open-source as they are integrated into enterprise environments.
Written by urgent.news from Computerworld's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
