Elicit tops BioASQ with 60.3% recall, beating Exa, Brave, Parallel and Perplexity
elicitorg · x · 2026-09-03
Elicit published a comparison against general-purpose web search APIs on the biomedical search benchmark BioASQ: Elicit achieved the highest recall at 60.3% for 50 papers, versus 40.3% for the next-best system.
The eval covered four web search APIs — Exa, Brave, Parallel and Perplexity. For fairness, all systems were restricted to academic sources (Exa via its publication filter; the others allow-listed to PubMed and PubMed Central), with each BioASQ question sent exactly as written. The post coincides with Elicit's recently launched API and MCP, giving developers programmatic access to its scientific search.
More from Apps
- every launches Compound Writing plugin that turns Claude Code into a writing coach — danshipper · 2026-09-03
- Amazon's AI assistant can now spot fake emails claiming to be from the company — The Verge AI · 2026-09-03
- Open-source Tangible Lesson toolkit makes science videos interruptible and hands-on — _lewtun · 2026-09-03
- umbrelOS 2.0 announced: home cloud OS public beta out now, launch set for September 22 — JosephJacks_ · 2026-09-03
- A one-line prompt to stop Claude from writing overwrought prose — iannuttall · 2026-09-03
- MagicPath plugin brings an infinite canvas to ChatGPT on web and desktop — jxnlco · 2026-09-03