GeekPark: Perplexity’s local AI gets a score boost from the cloud

According to GeekPark’s notes from Perplexity CEO Aravind Srinivas’s SGLang Summit talk, a fine-tuned Qwen 27B scored 59.6% locally on 89 tasks. Calling Opus 5 only when needed raised that to 73%. Local AI, with backup on speed dial. The notes also describe a Python SDK that lets agents build their own search pipelines.