A coalition of AI research labs released an open-source large language model that matches GPT-4’s performance on major benchmarks while running at a fraction of the cost. The model, called OpenLLM-200B, contains 200 billion parameters and can run on consumer hardware with just 24 gigabytes of RAM, compared to the enterprise-grade GPUs required for GPT-4. “This changes the competitive landscape entirely,” said Dr. Sarah Chen, an AI researcher at Stanford. “Companies that were locked out of advanced AI capabilities now have a viable path forward.” ...