Exa has launched Deep Max, a cutting-edge search API designed for large language model (LLM) agents, claiming it is up to 20 times faster than its closest competitors while maintaining high accuracy. The API employs a unique approach where each query initiates multiple concurrent searches, synthesizing results into a single, precise answer. This method, known as agentic search, allows LLMs to autonomously conduct multi-step searches and generate comprehensive responses. In internal benchmarks, Deep Max demonstrated superior performance across various tests. It achieved 90% accuracy in 64 seconds on the Deep Search QA test, significantly outperforming You.com Frontier and Parallel Ultra 8x, which took 5,908 and 1,703 seconds, respectively. On the FRAMES test, Deep Max reached 94% accuracy in just 11 seconds, while Parallel Ultra took 1,457 seconds. Additionally, in the HLE-Search test, Deep Max matched GPT-5.4's 80% accuracy but halved the response time. Currently, Deep Max is available to select customers, with broader release details yet to be announced.