Legal Retrieval Benchmark: three search APIs on 581 official-document requests
The Legal Retrieval Benchmark measures whether a search API returns the official document behind a legal research request, and whether it puts that document first. The public release runs the Firecrawl Government & Legal Index, Exa and Parallel on 581 requests, with Claude Code as a reference, built from what customers ask the index, with one frozen official answer each and no model judging the scores. The loop behind it has turned eight times inside Firecrawl; this is the first version published.
- Read the scores. gold@1, gold@10 and bad@10 for every engine, with 95% intervals, on the benchmark page.
- See the price of the official page. Index searches use no credits; the page charts each engine's score against its cost per request (list price for the engines, the agent's spend for the two Claude Code references) and against its median latency.
- Follow the loop. Every week, customer requests find the gaps in the index, the search team fixes them, and the next version measures the result.
- Try the index without an API key.
POST /v2/search/gov, thefirecrawl_gov_searchtool in the Firecrawl MCP server, orfirecrawl govin the CLI. - Rerun it. The 290 public requests, their answer key and the scoring code are on GitHub at firecrawl/benchmark-legal-retrieval.