arlen/benchOpen benchmarks for agentic consumers
SNAPSHOT web_search-2026-q3
11 JUN 2026 · BERKELEY, CA
Head-to-Head · Illustrative prototype

Exa vs Serper

Exa versus Serper for AI agents — hit@k accuracy, freshness lag, cost per verified-correct answer, and agent-readiness, scored against golden truth.

Answer firstillustrative

On the illustrative web_search-2026-q3 snapshot, Exa wins more head-to-head metrics (3 of 6). Exa leads on accuracy; compare cost per verified-correct answer and freshness below.

§ 01

Head to Head — Web Search

snapshot web_search-2026-q3
Metric ExaSerperWinner
hit@1 %71.266.0Exa
hit@5 %87.681.2Exa
fresh<30d %81.077.3Exa
retrievability h11.210.6Serper
cost/correct $$0.0064$0.0041Serper
p50 latency ms412355Serper
§ 02

Head to Head — Web Extraction

snapshot web_extraction-2026-q2
Metric ExaSerperWinner
fidelity 0-10.740.69Exa
JS gap Δ0.220.27Exa
block rate %8.89.5Exa
cost/correct $$0.0058$0.0027Serper
schema validity %84.0Exa
§ 03

Which Should an Agent Pick?

For accuracy-first agent workloads, the higher hit@5 vendor wins; for cost-sensitive high-volume use, prefer the lower cost-per-correct vendor. Both Exa and Serper should be evaluated on your own query mix — these are illustrative prototype figures.

Illustrative prototype. No verified vendor run has been published yet; every figure here is a placeholder and must not be cited as a measured result. Numbers are replaced when a snapshot’s first full run lands.