Search as Code directly turns the retrieval into Python code orchestration, maximizing parallelism—this approach is spot on.

CoinWorld News, Perplexity has fully launched Search as Code (SAC) in its agent API and self-developed computing products. The new architecture abandons multi-round serial inference calls, with the agent directly generating Python code that calls the Agentic Search SDK's retrieval primitives, executing thousands of searches and processing tasks in parallel within a single round. SAC allows Python to run in a secure sandbox, dynamically orchestrating parallel queries, deduplication, filtering, merging, and reordering. Benchmark evaluations show SAC achieving a score of 0.386, approximately 2.5 times higher than the next-best system, with 100% accuracy in real-world case tests.