Vendor Profile · Illustrative prototype
Tavily
Tavily on the arlen/bench leaderboards — agent web-search accuracy, extraction fidelity, freshness, cost per verified-correct answer, and agent-readiness, scored against golden truth.
Answer firstillustrative
Tavily ranks #4 of 4 for agent web-search accuracy at 49.4% hit@5; #3 of 4 for extraction fidelity at 0.62, and is not yet agent-ready in live onboarding trials.
hit@5
49.4%
web search
cost / correct
—
not yet measured
extraction fidelity
0.62
web extraction
agent-ready
No
onboarding harness
§ A
Web Search — Tavily
| Metric | Tavily | Leaderboard best | Direction |
|---|---|---|---|
| hit@1 % | 31.4 | Exa 62.2 | higher is better |
| hit@5 % | 49.4 | Exa 80.9 | higher is better |
| fresh<30d % | — | — | higher is better |
| retrievability h | — | — | lower is better |
| cost/correct $ | — | — | lower is better |
| p50 latency ms | — | — | lower is better |
§ B
Web Extraction — Tavily
| Metric | Tavily | Leaderboard best | Direction |
|---|---|---|---|
| fidelity 0-1 | 0.62 | Exa 0.74 | higher is better |
| phrase recall | 0.54 | Exa 0.67 | higher is better |
| boilerplate excl. | 0.68 | Exa 0.76 | higher is better |
| cost/correct $ | — | Exa $0.0047 | lower is better |
| coverage % | 98.7 | Jina 99.3 | higher is better |
§ C
Tavily vs Exa Tavily vs SerpAPI Tavily vs Brave Tavily vs Firecrawl Tavily vs Jina Tavily vs SerperCompare Tavily
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.