Firecrawl launches Research Index with /search/research API and 18% arXivQA recall gain

• Firecrawl's launch post says the new Research Index is a research-specific search index for AI and ML agents, and claims an 18% arXivQA recall lead over the next-best provider at similar cost.
• According to Firecrawl's corpus thread, the index covers 3M+ arXiv papers plus GitHub artifacts from top research repos, with daily refreshes.
• Firecrawl's toolset thread pitches the product as a full research loop, paper retrieval, claim verification against full text, and code lookup for implementation.
• Firecrawl's availability post says /search/research is live across the API, CLI, MCP, and SDKs, and works inside Codex, Claude Code, and Grok Build.
• Firecrawl's production-use thread adds an early customer claim from Aemon, where Firecrawl says internal recall scores beat Exa by 30% and Parallel by 250%.

You can jump straight to the product page, skim the corpus details, and watch the research workflow demo. The interesting part is not just another search endpoint. Firecrawl is packaging papers, code artifacts, and claim checking into one surface aimed squarely at research agents.

The retrieval pitch is breadth plus freshness. According to Firecrawl's corpus thread, the index includes all 3M+ arXiv papers and GitHub artifacts from top research repos, refreshed daily.

That targets a familiar failure mode for agentic research runs. Firecrawl's corpus thread argues general search providers often omit or mis-rank key papers, which forces humans back into manual source review.

Firecrawl is selling more than retrieval. Firecrawl's toolset thread says the Research Index ships with tools to:

• retrieve papers
• verify claims against full text
• pull code for implementation

That is a cleaner product boundary than a plain paper index. The bundle is meant to cover literature lookup, evidence checking, and repo discovery in the same run.

Availability is already spread across the usual agent surfaces. Firecrawl's availability post says the index is exposed through /search/research in the API, plus the CLI, MCP, and SDKs.

The same post says it works in:

• Codex
• Claude Code
• Grok Build

That matters mostly as a distribution signal. Firecrawl is not waiting for teams to adopt a new standalone interface first.

Firecrawl's top-line benchmark claim is an 18% recall advantage on arXivQA at similar cost, per the launch post. The company also says the index is already powering autonomous R&D at Aemon.

A second benchmark claim comes from customer usage. Firecrawl's production-use thread says Aemon's AI R&D engineer queries the index for papers, code, and technical discussions across the web, and that Aemon's internal benchmark scored Firecrawl 30% ahead of Exa and 250% ahead of Parallel on recall.

Those two numbers point at the same product thesis from different angles. One is a public benchmark, the other is an internal production comparison tied to a named research lab.