mims-harvard/ATHENA-R1-Qwen3-8B
mims-harvard/ATHENA-R1-Qwen3-8B is an 8 billion parameter AI agent developed by mims-harvard for treatment reasoning, built upon the Qwen3 architecture. It is specifically trained through reinforcement learning to perform multi-step reasoning over 212 biomedical tools, excelling at synthesizing evidence from authoritative sources to answer clinical questions. This model is optimized for complex biomedical decision support, demonstrating superior performance on drug and patient-specific treatment reasoning benchmarks compared to larger models.
Loading preview...
ATHENA-R1-Qwen3-8B: An AI Agent for Biomedical Treatment Reasoning
mims-harvard/ATHENA-R1-Qwen3-8B is an 8 billion parameter AI agent designed for advanced treatment reasoning in the biomedical domain. Developed by mims-harvard, this model is trained using reinforcement learning to navigate and utilize a vast universe of 212 biomedical tools, including FDA labeling, Open Targets, ChEMBL, and EuropePMC.
Key Capabilities
- Multi-step Reasoning: The model can identify necessary evidence, select appropriate tools, and integrate retrieved information into subsequent reasoning steps to answer complex clinical questions.
- Evidence Synthesis: It synthesizes evidence from authoritative biomedical sources to provide grounded, free-form answers.
- Tool Integration: Tool calls are managed through the ToolUniverse, allowing dynamic interaction with various data sources.
- High Performance: ATHENA-R1 significantly outperforms GPT-5 on specialized benchmarks, achieving 94.7% on DrugPC (drug reasoning) and 82.9% on TreatmentPC (patient-specific treatment).
Intended Use
This model is a research artifact primarily for treatment-reasoning research and decision support in biomedical contexts. It is not intended for direct patient care and should not be used as a medical device.