Biomni
Biomni: a general-purpose biomedical AI agent
Last verified:
What is Biomni?
Biomni is a general-purpose biomedical AI agent designed to autonomously execute a wide spectrum of research tasks across diverse biomedical subfields. Built by Stanford University researchers in collaboration with Genentech, the Arc Institute, University of Washington, Princeton University, and UCSF, it integrates large language model reasoning with retrieval-augmented planning and code-based execution to dynamically compose complex biomedical workflows without predefined templates.
The tool features access to 150 specialized tools, 59 biological databases, and 106 software packages systematically curated from 2,500+ bioRxiv papers across 25 biomedical subfields. Key capabilities include automating literature reviews, hypothesis generation, protocol design, bioinformatics analysis, clinical reasoning, CRISPR screen design, flow cytometry analysis, drug repurposing, causal gene prioritization, rare disease diagnosis, and microbiome analysis. It provides a no-code web interface at biomni.stanford.edu where scientists can immediately delegate research tasks.
Biomni is designed for biomedical scientists, researchers, and lab teams who want to enhance research productivity and generate testable hypotheses. It serves researchers working in genomics, microbiome, physiology, molecular biology, immunology, systems biology, bioengineering, and cell biology. The agent can handle complex multi-step research tasks spanning multiple domains, executing code in Python and R to perform real computational analyses.
The system uses Claude-powered AI (Claude 4 Sonnet) and features adaptive planning that iteratively refines plans as tasks execute. It includes a curated biological data lake with gene expression data from GTEx, cancer data from COSMIC, gene sets from MSigDB, drug binding affinities from BindingDB, disease associations from DisGeNET, and T-cell receptor data from McPAS-TCR. Biomni is an open-source initiative under Apache 2.0 license that invites the community to contribute tools, datasets, and benchmarks.
Biomni pricing
Pricing model: Freemium
Completely free web platform - Biomni is available at biomni.stanford.edu as a free web interface where biomedical scientists can immediately delegate research tasks at no cost. No credit card required. The web platform supports concurrent tasks with scalable infrastructure and compute, and includes high performance computing CPU and GPU jobs. For API access outside the web platform (Amazon Bedrock, Google Cloud), pricing is $3 per million input tokens and $15 per million output tokens. Biomni is an open-source initiative under Apache 2.0 license, though certain integrated tools, databases, or software may carry more restrictive commercial licenses requiring careful review before commercial use.
Biomni pros
- Completely free web platform for scientists with no credit card required
- Access to 150 specialized biomedical tools in one unified environment
- 59 biological databases with natural language query interfaces
- 106 software packages for bioinformatics analysis
- Autonomously executes complex multi-step research workflows
- No predefined templates needed - dynamically composes workflows
- Strong benchmark performance beating human experts on LAB-Bench (74.4% DbQA, 81.9% SeqQA)
- Outperforms base LLMs by 402.3% on HLE benchmark across 14 subfields
- Handles 25 different biomedical subfields from CRISPR to microbiome
- Complete no-code web interface - no programming required
- Generates testable hypotheses from data analysis
- Can complete tasks 100x faster than humans (35 min vs 3 weeks example)
- Open-source under Apache 2.0 license with community contributions welcome
- Validated in wet lab matching >5-year expert performance in blinded test
- Adaptive planning refines plans during execution when steps fail
- Access to curated biological data lake with 11GB of biomedical data
- Supports concurrent tasks with scalable infrastructure and compute
Biomni cons
- Executes LLM-generated code with full system privileges - security risk
- Requires isolated/sandboxed environments for production use
- Agent can access files, network, and system commands - be careful with sensitive data
- Release frozen as of April 15 2025 differs from current web platform
- Certain integrated tools/databases may have restrictive commercial licenses beyond Apache 2.0
- Still underperforms human experts roughly two-thirds of the time on BioML-bench (34.45th percentile)
- May run out of memory trying to copy large image files
- Can have silent code errors preventing final predictions from saving
- Environment setup is massive requiring complex conda/pip installation for local use
- API costs apply if using outside free web platform ($3 per million input tokens, $15 per million output tokens)
Frequently asked questions about Biomni
What is Biomni?
Biomni is the first general-purpose biomedical AI agent designed to autonomously execute a wide spectrum of research tasks across diverse biomedical subfields. It integrates large language model reasoning with retrieval-augmented planning and code-based execution, enabling it to dynamically compose and carry out complex biomedical workflows entirely without relying on predefined templates or rigid task flows. Built on Biomni-E1 (unified biomedical environment with 150 tools, 59 databases, 106 software packages) and Biomni-A1 (generalist agent with retrieval, planning, and code as action).
Who should use Biomni?
Biomni is designed for biomedical scientists, researchers, and lab teams who want to dramatically enhance research productivity and generate testable hypotheses. It serves researchers working across 25 biomedical subfields including genomics, microbiome, physiology, molecular biology, immunology, systems biology, bioengineering, and cell biology. The no-code web interface makes it accessible to scientists without programming expertise who want to delegate research tasks to an AI agent.
What tasks can Biomni automate?
Biomni automates literature reviews, hypothesis generation, protocol design, bioinformatics analysis, clinical reasoning, CRISPR screen design, flow cytometry analysis, drug repurposing, causal gene prioritization, rare disease diagnosis, and microbiome analysis. It can perform activities like querying PubMed for papers, designing knockout sgRNA, analyzing Cas9 mutation outcomes, performing flux balance analysis, querying Uniprot/AlphaFold/KEGG/Ensembl/GWAS Catalog, and executing complex multi-step workflows spanning multiple domains.
How do I access Biomni?
Biomni is available through a completely free no-code web interface at biomni.stanford.edu where scientists can immediately delegate research tasks. For local installation, you can use the setup.sh script to setup the environment, activate conda environment biomni_e1, then install the biomni pip package. API access is also available through Amazon Bedrock and Google Cloud for programmatic use.
Is Biomni free to use?
Yes, Biomni is completely free through its web platform at biomni.stanford.edu. Scientists around the world can delegate research tasks to Biomni at no cost with no credit card required. The free web platform includes private secure workspace, scalable infrastructure with concurrent tasks, and HPC CPU/GPU job support. Biomni is also an open-source initiative under Apache 2.0 license.
What benchmarks has Biomni passed?
On LAB-Bench, Biomni achieved 74.4% accuracy in DbQA and 81.9% in SeqQA, outperforming human experts who scored 74.7% and 78.8% respectively. On the HLE benchmark covering 14 subfields, Biomni scored 17.3%, outperforming base LLMs by 402.3%, coding agents by 43.0%, and its ablated variant by 20.4%. In a real-world test, Biomni completed wearable bioinformatics analysis across 458 messy files in 35 minutes versus 3 weeks by a human.
What databases does Biomni access?
Biomni provides access to 59 biological databases including UniProt, AlphaFold, KEGG, Ensembl, GWAS Catalog, Reactome, GTOPDB, and OpenTarget. The curated data lake includes GTEx tissue gene expression (TPM across human tissues), COSMIC cancer gene expression, MSigDB curated gene sets and cell type signatures, BindingDB protein-small molecule binding affinities, DisGeNET gene-disease associations, and McPAS-TCR T-cell receptor sequences with specificity data.
Can Biomni contribute to actual research discoveries?
Yes, Biomni has demonstrated real research contributions. It designed a molecular cloning experiment that was validated in wet lab, matching the performance of a >5-year expert in a blinded test. It uncovered novel transcription factor hypotheses driving skeletal lineage regulation. The tool is designed to generate testable hypotheses and help scientists co-create the next era of biomedical discoveries through autonomous task execution.
Is Biomni open source?
Biomni itself is Apache 2.0-licensed and is an open-science initiative that welcomes community contributions. The code is available at github.com/snap-stanford/biomni and will be fully open-sourced soon. However, certain integrated tools, databases, or software may carry more restrictive commercial licenses, so users should review each component carefully before commercial use. Contributors can add new tools, datasets, software, benchmarks, and tutorials.
What are the security concerns with Biomni?
Biomni executes LLM-generated code with full system privileges, which presents security risks. For production use, it should be used in isolated/sandboxed environments. The agent can access files, network, and system commands, so users must be careful with sensitive data or credentials. This is an important consideration when deploying Biomni for real research workflows involving confidential or proprietary information.