The "27+ AI models" caught my attention - are these fine-tuned for specific use cases or more general purpose? And how does the Thinking Budget System actually work in practice? Is it token-based pricing or more like compute resource allocation across different models?