title: "Stanford Paper2Agent hits 91.2% accuracy turning research papers into " slug: "stanford-paper2agent-hits-912-accuracy-turning-research-papers-into-live-ai-agen" published: "2026-09-23" beat: "Research" tags: ["Research", "Tools"] creator: "Agentry Newsroom" editor: "Susanne Sperling, Editor — Human in the Loop" tools: ["Claude (Anthropic)", "Perplexity Sonar"] creativeWorkStatus: "verified" dateReviewed: "2026-09-23" aiActArticle50: "compliant" humanView: "https://agentry.news/research/stanford-paper2agent-hits-912-accuracy-turning-research-papers-into-live-ai-agen" agentView: "https://agentry.news/agent/stanford-paper2agent-hits-912-accuracy-turning-research-papers-into-live-ai-agen"
Stanford researchers released Paper2Agent on September 16, 2026, a system that automatically converts research papers, code, and datasets into interactive AI agents. The team reported 91.2% accuracy o
Drafted by an AI agent. Verified by Susanne Sperling, Editor — Human in the Loop. AI policy.
Stanford researchers published Paper2Agent in Nature on September 16, 2026, describing a system that automatically transforms research papers, supplementary code, datasets, and materials into interactive AI agents that can reproduce published results and execute novel queries Marktechpost.
The system achieved 91.2% average accuracy across a 300-question benchmark, with secondary evaluations confirming strong performance on paper-derived tasks and non-biology domains Marktechpost. In a targeted assessment, the team evaluated Paper2Agent on 100 computational biology papers sourced from bioRxiv, successfully converting 74 into functional agents. Of the 599 tools the system proposed across these conversions, 593 passed validation testing Digital Applied.
The researchers demonstrated practical utility by constructing a working genomics agent from a published paper in approximately 45 minutes at a reported computational cost of $14. When tested on 15 novel queries, the agent achieved either 100% accuracy or near-perfect performance in independent evaluations IEEE Spectrum.
Paper2Agent addresses a friction point in research reproducibility. Rather than requiring researchers to manually implement code and integrate datasets, the system ingests published materials—papers, code repositories, supplementary files—and outputs an agent capable of:
• Reproducing the original study's results
• Running inference on new input data
• Answering queries derived from the paper's methods and findings
• Validating proposed computational tools against real workflows
The system's ability to agentify biology papers at a 74% success rate (out of 100 papers tested) indicates the approach scales beyond proof-of-concept toy examples Digital Applied.
Paper2Agent ships as a concrete research tool that reduces the engineering burden of converting academic work into production-ready agents. At $14 per agent for an hour of compute, the cost structure is measurable and reproducible—metrics absent from most agent-platform announcements. The benchmark results (91.2% accuracy, 593/599 tool validations) provide quantified proof that the system handles real research artifacts, not sanitized datasets.
For enterprises and research institutions seeking to operationalize published findings, the system offers a documented, repeatable path from paper to deployment. The September 16 Nature publication in a top-tier venue validates the research contribution, while the concrete benchmarks and cost figures provide the verification rigor required to move beyond roadmap hype.