8 web data APIs, every track they ran
Each cell is tasks done with the provider as the agent's only tool, and the cost of each finished task. Bold marks the most tasks done on a track, which on most tracks is not a gap big enough to rule out chance. A dash means the provider didn't run that track.
| Provider | Scrape · 60 tasks | Search · 60 tasks | Map · 15 tasks | Crawl · 10 tasks |
|---|---|---|---|---|
| Context.devcontent API | 92%0.72¢ | 75%1.82¢ | 47%1.84¢ | 60%7.58¢ |
| Firecrawlcontent API | 90%1.22¢ | 85%3.04¢ | 40%3.35¢ | 50%14.2¢ |
| Browserbasebrowser infrastructure | 82%1.12¢ | 38%5.97¢ | not run | not run |
| Linkupcontent API | 90%0.60¢ | 83%2.55¢ | not run | not run |
| Scrapflyscraping infrastructure | 88%0.61¢ | not run | not run | 70%2.88¢ |
| TinyFishcontent API | 90%0.43¢ | 73%1.31¢ | not run | not run |
| Jinacontent API | 87%0.57¢ | not run | not run | not run |
| Spidercontent API | 20%2.48¢ | not run | not run | not run |
Not in this table
Exa, Tavily, Bright Data and Perplexity were not tested, so there are no numbers for them here. Parallel was tested on search, but its customer terms require written consent to publish benchmark results, so its results are left out. Want a head-to-head? See the comparisons.