Perplexity’s Search as Code (SaC) research published June 1, 2026. The flagship demo is a security research task: identify and characterize 200+ high-severity CVEs from 2023–2025. SaC scored 100% accuracy and used 85.1% fewer tokens (42.9K vs 288.7K).
Systems tested: SaC on Perplexity Agent API (GPT 5.5, high reasoning), OpenAI Responses API with web_search and code_interpreter, Anthropic Managed Agents (Opus 4.7), Exa, and Parallel.
The model is the control plane; the compute sandbox handles deterministic work (control flow, batching, retries, filtering, joins). The Agentic Search SDK exposes primitives so one inference turn can drive thousands of operations.
SaC ships as default in Perplexity Computer and is available in the Agent API. The Agentic Search SDK and retrieval infrastructure are internal to Perplexity.