The main update is a new search approach called Search as Code.
Simply put: now search can be "assembled like a constructor" directly in the code, and the agent itself chooses the best way to find information for a specific task.
What this gives:
more accurate answers (even for complex research)
lower costs (up to −85% compared to other solutions)
better performance for both quick queries and deep analysis
Example:
The agent was tasked with finding vulnerabilities (CVE) and official patches — it delivered 100% accurate results, while other systems (OpenAI, Anthropic, Exa) — less than 25%.
#AItools #Perplexity
🐯 Time2Future | AI Media