← Home
Benchmark report

State of Web-Access APIs for AI Agents (2026)

We benchmarked 22 web-access APIs on one corpus across 13 capabilities. Firecrawl won the most; no provider won them all. The merit-only ranking.

Updated 2026-06-30 · Auxiliar

The Auxiliar curator fleet ran the major web-access APIs head-to-head on the same corpus, scoring each on quality, latency, cost per success and error rate — per capability. This is the merit-only result: no house provider, no thumb on the scale, the same tests for everyone. Here’s what 22 providers across 13 capabilities actually look like when you measure them.

Headline findings

  • Firecrawl won 5 of 13 capabilities — more than any other provider — leading scrape, crawl, screenshot, PDF parsing and change detection.
  • No provider wins everything. The best SERP provider, the best AI-search index, the best extractor and the best unblocker are four different companies. Standardizing on one vendor means using its weakest verb somewhere.
  • Anti-bot is solved — for some. Firecrawl, Scrapfly cleared 100% of the anti-bot vendors in the corpus; others failed outright on the same targets.
  • Price spans orders of magnitude. On SERP, Serper came in cheapest at $0.0003 / call — a fraction of the premium incumbents for comparable quality.

Winners by capability

Every score is a composite (0–10) from the same corpus. Each links through to the full scorecard.

CapabilityWinnerScoreRunner-upMeasured on
Web search (RAG grounding)Jina7.69/10You.com (7.41)Recall@10
Scrape to markdownFirecrawl9.81/10Scrapfly (9.48)Anti-bot bypass
SERP / Google resultsSerper9.56/10SearchAPI.io (9.02)Quality
Cited answers & researchExa7.9/10You.com (7.79)Correctness
Site crawlingFirecrawl8.61/10Spider (8.54)Coverage
AI / schema extractionScrapfly8.96/10Firecrawl (8.81)Field accuracy
Rule-based (CSS) extractionScrapingBee9.23/10Oxylabs (8.52)Field accuracy
ScreenshotsFirecrawl9.5/10Zyte (9.04)Valid image
Structured domain scrapingOxylabs8.74/10Apify (8.23)Accuracy
PDF / document parsingFirecrawl8.82/10Jina (8.43)Text accuracy
Declarative browser actionsScrapingBee8.72/10Zyte (8.5)Task success
Natural-language browser agentsOxylabs7.99/10Firecrawl (5.66)Task success
Change detectionFirecrawl8.33/10Class. accuracy

The standouts

Why we published this

The web-access market is dozens of providers all claiming to be fastest, cheapest and most reliable. Almost none publish head-to-head numbers — and the few “benchmarks” that exist are run by vendors grading their own homework. Auxiliar has no reason to favor one provider over another: it resells all of them on one key. So the ranking above is the honest one, and you can act on it directly — every provider here is reachable through a single Auxiliar API key, so you can route each job to the winner (or fall back between them) without a new signup.

Methodology

Each provider was exercised on identical inputs per capability (“verb”) — the same URLs to scrape, the same queries to search, the same documents to parse. We scored verb-specific quality (e.g. anti-bot bypass rate, markdown cleanliness, recall, field accuracy), latency p50, cost per successful result, and error rate, then combined them into a composite. Providers that couldn’t be scored on a synchronous request/response corpus (e.g. stateful browser sessions) are noted on their scorecards rather than ranked. Numbers are refreshed as the fleet re-runs; see any provider’s page for its latest detail.

One key. Every provider on this page.

Stop juggling signups and invoices. One Auxiliar API key calls all of them — upstream keys injected server-side, usage billed to a single balance. Swap the base URL and go.

curl https://api.auxiliar.ai/serper/search \
  -H "Authorization: Bearer $AUXILIAR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"q": "latest ai agent news"}'

Keep building