Choosing a search API for an AI agent is no longer just about latency. Artificial Analysis has released the Search Index, a standard that evaluates providers on result quality, total cost, and execution speed. The benchmark tests seven services—Parallel, Exa, Firecrawl, You.com, Tavily, Keenable, and Brave—keeping the model (GPT-5.6 Luna) and agentic setup constant while varying only the underlying search engine.

Psst!Add AlexTech.ai asPreferred Source on Google

Parallel leads with a total cost advantage

The results show that search quality directly impacts agent economics: without web access, the model scores 33 points; with active search, scores range from 65 to 75. Parallel takes first place with a score of 75, followed by Exa (74) and Firecrawl (73). A crucial data point emerges from token analysis: with Parallel Search in advanced mode, token usage drops by over 40% compared to the basic version. Although per-query costs increase, overall efficiency brings total spend down to $0.084 versus $0.11 for the basic version.

Per-query speed does not guarantee faster task completion

The benchmark challenges the assumption that a faster response per query translates to quicker task completion. Parallel Search (turbo) records the shortest response time (0.51 seconds vs. 1.03 for basic), but its lower quality (67 points vs. 73) forces the agent to repeat more passes. Total time per task remains substantially unchanged, confirming that retrieval quality is the dominant factor for end-to-end efficiency.

Transparent methodology and openness to new providers

The index combines three equally weighted benchmarks: DeepSearchQA (900 research questions), a BrowseComp subset (200 hard-to-find facts), and AA-Omniscience (600 questions across six knowledge domains). The open-source Stirrup framework, developed by Artificial Analysis, manages execution with 25 runs per task. The methodology is public and other providers can apply for inclusion, consolidating an objective reference point in a market that has until now been fragmented and lacking comparative standards.