Sourcery – Benchmark for search APIs on the sources they return
Independent benchmark that scores search APIs by the sources they actually return.
Sourcery grades eight search APIs — Perplexity, Exa, Parallel, Brave, Firecrawl, Tavily and others — not on the model's final answer but on the quality of the sources each one returns, judged by an LLM against 7,496 unique question-page pairs after deduplication. It's self-funded (about $40.08 spent across all eight providers, no sponsor saw it first) and every rating is checkable in an explorer that shows the judge's own justification. Perplexity comes out best overall on answer completeness but returns excerpts rather than whole pages; Exa is rated best for feeding agents structured, link-free text.
What holds up
- +Self-funded and independent — no provider paid for or previewed the results.
- +Every rating is checkable: the explorer shows the judge's own justification.
- +Deduplicated 12,954 pairs to 7,496 unique before grading, cutting judge calls 42%.
Mind the limits
- −Results truncated to 1,600 characters before judging, which could reorder providers.
- −One person's run on about $40 of credits — a small sample, not an ongoing benchmark.
Featured here? Take the badge
Put it on your site — it links back to this review. Free for every listed product, always.
<a href="https://stillworks.dev/products/p/sourcery-benchmark-for-search-apis-on-the-sources-they-retur/"><img src="https://stillworks.dev/badge/sourcery-benchmark-for-search-apis-on-the-sources-they-retur.svg" alt="Picked by StillWorks" width="250" height="54"></a>