We just deployed a custom 12b model that we fine-tuned: - better performance than GPT 4.1 - half the latency - half the cost - hosted on commodity hardware OpenAI loses $40k/mo in revenue. Foundational labs are in trouble if this keeps working.
12,97K