Hunt Globally: Deep Research AI Agents for Drug Asset Scouting in Investing, Business Development, and Search & Evaluation

Dimensions overall score 8.0

• The Bioptic Agent achieves 79.7% F1 score in drug asset scouting, significantly outperforming state-of-the-art AI systems like Claude Opus 4.6 (56.2%), Gemini 3 Pro + Deep Research (50.6%), GPT-5.2 Pro (46.6%), Perplexity Deep Research (44.2%), and Exa Websets (26.9%).result 95%
  Evidencepartial

  Bioptic Agent achieves 79.7% F1 versus 56.2% (Claude Opus 4.6), 50.6% (Gemini 3 Pro + Deep Research), 46.6% (GPT-5.2 Pro), 44.2% (Perplexity Deep Research), and 26.9% (Exa Websets).

  Implicationpartial

  Directly stated in abstract with specific numeric results comparing multiple systems

  Verificationpartial

  partial

• More than 85% of patent filings for drug assets originate outside the United States, with China accounting for nearly half of the global total.market 90%
  Evidencepartial

  Recent data suggests >85% of patent filings originate outside the U.S., with China accounting for nearly half of the global total

  Implicationpartial

  Explicitly stated in abstract with specific percentage data

  Verificationpartial

  partial

• China accounts for approximately 30% of global drug development, spanning over 1,200 novel candidates.market 90%
  Evidencepartial

  Industry estimates put China at ~30% of global drug development, spanning 1,200+ novel candidates.

  Implicationpartial

  Directly stated in abstract with specific numeric data

  Verificationpartial

  partial

• The system's performance improves steeply with additional compute, supporting the view that more compute yields better results.technical 85%
  Evidencepartial

  Performance improves steeply with additional compute, supporting the view that more compute yields better results.

  Implicationpartial

  Directly stated in abstract as a conclusion from the research

  Verificationpartial

  partial

• The Bioptic Agent uses a tree-based self-learning architecture aimed at complete, non-hallucinated scouting of drug assets.method 85%
  Evidencepartial

  We propose a benchmarking methodology for drug asset scouting and a tuned, tree-based self-learning Bioptic Agent aimed at complete, non-hallucinated scouting.

  Implicationpartial

  Directly stated in abstract describing the method

  Verificationpartial

  partial

• The benchmark was constructed using complex user queries paired with ground-truth assets that are largely outside U.S.-centric radar.method 80%
  Evidencepartial

  We construct a challenging completeness benchmark using a multilingual multi-agent pipeline: complex user queries paired with ground-truth assets that are largely outside U.S.-centric radar.

  Implicationpartial

  Directly stated in abstract describing the benchmarking methodology

  Verificationpartial

  partial

• The system's performance may degrade with languages not included in its training scope or with assets disclosed in less accessible sources.limitation 85%
  Evidencepartial

  The system's performance may degrade with languages not included in its training scope or with assets disclosed in less accessible sources.

  Implicationpartial

  Directly stated in analysis section as a caveat

  Verificationpartial

  partial

• Today's Deep Research AI agents still lag human experts in achieving high-recall discovery across heterogeneous, multilingual sources without hallucinations.limitation 80%
  Evidencepartial

  Yet today's Deep Research AI agents still lag human experts in achieving high-recall discovery across heterogeneous, multilingual sources without hallucinations.

  Implicationpartial

  Directly stated in abstract as a problem statement

  Verificationpartial

  partial

Competitive landscape

AI agents for comprehensive global drug asset scouting in biopharma investments.