How to run Open Interpreter with the latest models (DeepSeek V4, Claude Sonnet 4.5, GPT-5.5) for 85-95% less β using Lynkr for automatic tier routing and prompt caching.