Parallel Search scores 83 on Artificial Analysis Search Index

Summary

Parallel Search has achieved a score of 83 on the Artificial Analysis Search Index, matching the performance of Keenable Search (pro) and demonstrating significant improvements in their benchmark results from August to November, where their advanced mode increased from 75 to 83 and the fast mode from 73 to 82. This advancement highlights a trend in search tool performance evaluations, as multiple providers continue to show gains in independent assessments, with AI agents increasingly integrating language models with specialized search APIs for more effective handling of complex tasks. Additionally, while the fast mode is priced at $0.014 per task, making it the second lowest option available, the advanced mode costs slightly more at $0.030 per task.

Analysis

Stirrup: Stirrup is an agent evaluation harness designed to test search tools across large sets of tasks. It integrates external search APIs into model-driven workflows for standardized performance measurement. In the news, it served as the testing framework for Parallel Search modes on 1700 search-focused tasks. DeepSearchQA: DeepSearchQA is a specialized benchmark focused on evaluating search-augmented question answering capabilities. It measures F1 scores on search-intensive tasks to assess tool effectiveness. In the news, it is highlighted as the strongest area for Parallel Search, where both modes achieved leading results among evaluated providers. GPT-5.6 Luna: GPT-5.6 Luna is an advanced language model variant used as the core reasoning engine in agent setups. The medium configuration balances capability and efficiency for tool-using workflows. In the news, it powers the Stirrup agent harness during the Parallel Search benchmarking runs. Keenable Search: Keenable Search is a competing AI-powered search service with a pro tier that emphasizes dynamic and high-performance query handling. It serves as a benchmark reference point for other search tools in independent evaluations. In the news, it is directly compared to Parallel Search, matching its advanced mode score on the Artificial Analysis Search Index. Parallel Search: Parallel Search is a web search API offered with distinct fast and advanced operational modes, designed for integration as a tool in AI agent systems. It is developed by the team behind @p0 and focuses on objective-driven search queries to support complex tasks. In the news, Parallel Search is the primary subject of recent benchmarking, where both modes achieved top rankings on the Artificial Analysis Search Index and related evaluations when used within the Stirrup harness. Tool Integration: AI agents increasingly pair language models with specialized search APIs to handle complex, multi-query tasks. Benchmarking Trends: Search tool performance evaluations show continued gains across multiple providers in independent assessments.

Categories

aitechmachine_learning
View Original Tweet