Open Source AI Communities Are Crushing Big Tech at Its Own Game The Performance Gap That Big Tech Tried to Hide Meta released Llama 3 70B in April 2024 with an MMLU score of 86.0 percent. That sits within 0.4 points of GPT-4 Turbo while running on hardware costing under 0,000 per node instead of the multi-million-dollar clusters Google keeps private. The open weights let any lab replicate the...