Small AI model achieves frontier LLM performance with agentic workflow

Researchers have developed the DeepLens Diagnosis Agent, a novel five-stage pipeline designed to enhance the diagnostic reasoning of smaller, specialized AI models. This agentic workflow, which incorporates retrieval-augmented generation (RAG) and structured clinical extraction, significantly boosted the performance of the JSL Medical Small 7B v2 model. On the DiagnosisArena benchmark, the agent achieved 60.14% top-1 diagnostic accuracy, a substantial improvement over the base model's 23.99%. Furthermore, the agent proved more cost-effective and faster than larger frontier models like Claude Sonnet 4.5 and Gemini-3.1 Pro, while still outperforming them in accuracy. AI

IMPACT Demonstrates that specialized agentic workflows can enable smaller models to rival frontier LLMs in complex reasoning tasks, potentially lowering costs and increasing accessibility.

RANK_REASON The cluster contains a research paper detailing a novel AI methodology and benchmark results. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Claude Sonnet 4.5
• DeepLens Diagnosis Agent
• DiagnosisArena
• Gemini-3.1 Pro
• Hugging Face
• JSL Medical Small 7B v2

• model release
• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →