Biomni-Eval

Generalist biomedical agent evaluation suite across genetics, pharmacology, pathology, and clinical microbiology.27 Oct 2026

One row per model and compatible score measure. Mean = equal-weight arithmetic average of imported reports. Distinct configurations reported for the same model in a source table remain separate. Reports can overlap; the range is not a confidence interval. Unbundle to inspect each observation.
		Evidence	What the result shows
Biomni-R0-32B →	0.8 / 1 1 report	Biomni-Eval · Biomni-Eval1 · CRISPR delivery · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison (tied). Source notes Source cell: 0.8
Biomni-R0-8B →	0.8 / 1 1 report	Biomni-Eval · Biomni-Eval1 · CRISPR delivery · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison (tied). Source notes Source cell: 0.8
Qwen3-32B →Qwen	0.7 / 1 1 report	Biomni-Eval · Biomni-Eval1 · CRISPR delivery · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#3 of 6 in this displayed comparison; 0.1 / 1 below the leader. Source notes Source cell: 0.7
Qwen3-8B →Qwen	0.6 / 1 1 report	Biomni-Eval · Biomni-Eval1 · CRISPR delivery · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#4 of 6 in this displayed comparison; 0.2 / 1 below the leader. Source notes Source cell: 0.6
Claude 4 Sonnet →Anthropic	0.35 / 1 1 report	Biomni-Eval · Biomni-Eval1 · CRISPR delivery · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#5 of 6 in this displayed comparison; 0.45 / 1 below the leader. Source notes Source cell: 0.35
GPT-5 →OpenAI	0 / 1 1 report	Biomni-Eval · Biomni-Eval1 · CRISPR delivery · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#6 of 6 in this displayed comparison; 0.8 / 1 below the leader. Source notes Source cell: 0.0
GPT-5 →OpenAI	0.62 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · catalog · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison. Source notes Source cell: 0.62
Biomni-R0-32B →	0.58 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · catalog · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#2 of 6 in this displayed comparison; 0.04 / 1 below the leader. Source notes Source cell: 0.58
Biomni-R0-8B →	0.56 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · catalog · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#3 of 6 in this displayed comparison; 0.06 / 1 below the leader. Source notes Source cell: 0.56
Claude 4 Sonnet →Anthropic	0.5 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · catalog · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#4 of 6 in this displayed comparison; 0.12 / 1 below the leader. Source notes Source cell: 0.5
Qwen3-8B →Qwen	0.22 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · catalog · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#5 of 6 in this displayed comparison; 0.4 / 1 below the leader. Source notes Source cell: 0.22
Qwen3-32B →Qwen	0.2 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · catalog · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#6 of 6 in this displayed comparison; 0.42 / 1 below the leader. Source notes Source cell: 0.2
Biomni-R0-32B →	0.86 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · OpenTargets · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison. Source notes Source cell: 0.86
Biomni-R0-8B →	0.78 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · OpenTargets · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#2 of 6 in this displayed comparison; 0.08 / 1 below the leader. Source notes Source cell: 0.78
Claude 4 Sonnet →Anthropic	0.78 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · OpenTargets · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#2 of 6 in this displayed comparison; 0.08 / 1 below the leader. Source notes Source cell: 0.78
Qwen3-8B →Qwen	0.76 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · OpenTargets · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#4 of 6 in this displayed comparison; 0.1 / 1 below the leader. Source notes Source cell: 0.76
GPT-5 →OpenAI	0.74 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · OpenTargets · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#5 of 6 in this displayed comparison; 0.12 / 1 below the leader. Source notes Source cell: 0.74
Qwen3-32B →Qwen	0.7 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · OpenTargets · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#6 of 6 in this displayed comparison; 0.16 / 1 below the leader. Source notes Source cell: 0.7
Biomni-R0-32B →	0.8 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · PharmaProjects · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison. Source notes Source cell: 0.8
Qwen3-8B →Qwen	0.72 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · PharmaProjects · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#2 of 6 in this displayed comparison; 0.08 / 1 below the leader. Source notes Source cell: 0.72
Biomni-R0-8B →	0.7 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · PharmaProjects · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#3 of 6 in this displayed comparison; 0.1 / 1 below the leader. Source notes Source cell: 0.7
Claude 4 Sonnet →Anthropic	0.68 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · PharmaProjects · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#4 of 6 in this displayed comparison; 0.12 / 1 below the leader. Source notes Source cell: 0.68
GPT-5 →OpenAI	0.62 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · PharmaProjects · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#5 of 6 in this displayed comparison; 0.18 / 1 below the leader. Source notes Source cell: 0.62
Qwen3-32B →Qwen	0.62 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS causal gene · PharmaProjects · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#5 of 6 in this displayed comparison; 0.18 / 1 below the leader. Source notes Source cell: 0.62
Biomni-R0-32B →	0.74 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS variant prioritization · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison. Source notes Source cell: 0.74
GPT-5 →OpenAI	0.63 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS variant prioritization · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#2 of 6 in this displayed comparison; 0.11 / 1 below the leader. Source notes Source cell: 0.63
Claude 4 Sonnet →Anthropic	0.6 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS variant prioritization · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#3 of 6 in this displayed comparison; 0.14 / 1 below the leader. Source notes Source cell: 0.6
Biomni-R0-8B →	0.35 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS variant prioritization · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#4 of 6 in this displayed comparison; 0.39 / 1 below the leader. Source notes Source cell: 0.35
Qwen3-32B →Qwen	0.16 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS variant prioritization · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#5 of 6 in this displayed comparison; 0.58 / 1 below the leader. Source notes Source cell: 0.16
Qwen3-8B →Qwen	0.09 / 1 1 report	Biomni-Eval · Biomni-Eval1 · GWAS variant prioritization · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#6 of 6 in this displayed comparison; 0.65 / 1 below the leader. Source notes Source cell: 0.09
Biomni-R0-32B →	0.74 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench DbQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison. Source notes Source cell: 0.74
Biomni-R0-8B →	0.58 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench DbQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#2 of 6 in this displayed comparison; 0.16 / 1 below the leader. Source notes Source cell: 0.58
GPT-5 →OpenAI	0.54 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench DbQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#3 of 6 in this displayed comparison; 0.2 / 1 below the leader. Source notes Source cell: 0.54
Claude 4 Sonnet →Anthropic	0.52 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench DbQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#4 of 6 in this displayed comparison; 0.22 / 1 below the leader. Source notes Source cell: 0.52
Qwen3-32B →Qwen	0.2 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench DbQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#5 of 6 in this displayed comparison; 0.54 / 1 below the leader. Source notes Source cell: 0.2
Qwen3-8B →Qwen	0.16 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench DbQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#6 of 6 in this displayed comparison; 0.58 / 1 below the leader. Source notes Source cell: 0.16
Claude 4 Sonnet →Anthropic	0.86 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench SeqQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison. Source notes Source cell: 0.86
Biomni-R0-32B →	0.8 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench SeqQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#2 of 6 in this displayed comparison; 0.06 / 1 below the leader. Source notes Source cell: 0.8
GPT-5 →OpenAI	0.72 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench SeqQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#3 of 6 in this displayed comparison; 0.14 / 1 below the leader. Source notes Source cell: 0.72
Biomni-R0-8B →	0.7 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench SeqQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#4 of 6 in this displayed comparison; 0.16 / 1 below the leader. Source notes Source cell: 0.7
Qwen3-32B →Qwen	0.52 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench SeqQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#5 of 6 in this displayed comparison; 0.34 / 1 below the leader. Source notes Source cell: 0.52
Qwen3-8B →Qwen	0.28 / 1 1 report	Biomni-Eval · Biomni-Eval1 · LAB-Bench SeqQA · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#6 of 6 in this displayed comparison; 0.58 / 1 below the leader. Source notes Source cell: 0.28
Biomni-R0-32B →	0.67 / 1 1 report	Biomni-Eval · Biomni-Eval1 · mean task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison. Source notes Source cell: 0.669
Biomni-R0-8B →	0.59 / 1 1 report	Biomni-Eval · Biomni-Eval1 · mean task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#2 of 6 in this displayed comparison; 0.08 / 1 below the leader. Source notes Source cell: 0.588
Claude 4 Sonnet →Anthropic	0.56 / 1 1 report	Biomni-Eval · Biomni-Eval1 · mean task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#3 of 6 in this displayed comparison; 0.11 / 1 below the leader. Source notes Source cell: 0.557
GPT-5 →OpenAI	0.55 / 1 1 report	Biomni-Eval · Biomni-Eval1 · mean task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#4 of 6 in this displayed comparison; 0.12 / 1 below the leader. Source notes Source cell: 0.546
Qwen3-32B →Qwen	0.35 / 1 1 report	Biomni-Eval · Biomni-Eval1 · mean task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#5 of 6 in this displayed comparison; 0.32 / 1 below the leader. Source notes Source cell: 0.346
Qwen3-8B →Qwen	0.32 / 1 1 report	Biomni-Eval · Biomni-Eval1 · mean task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#6 of 6 in this displayed comparison; 0.35 / 1 below the leader. Source notes Source cell: 0.318
GPT-5 →OpenAI	0.4 / 1 1 report	Biomni-Eval · Biomni-Eval1 · Patient gene detection · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	Leads this displayed comparison. Source notes Source cell: 0.4
Biomni-R0-32B →	0.34 / 1 1 report	Biomni-Eval · Biomni-Eval1 · Patient gene detection · task reward[1]Biomni-R0 · A1 scaffold and E1 environment	#2 of 6 in this displayed comparison; 0.06 / 1 below the leader. Source notes Source cell: 0.34

Sources

Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
Ryan Li, Kexin Huang, Shiyi Cao, Yuanhao Qu, Jure Leskovec · 2025
Dates & evaluation details
Biomni-Eval · Biomni-Eval1 · mean task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact labeled means across ten biomedical tasks. This is an equal-weight mean of task rewards, not necessarily micro-accuracy across all 433 instances. All six systems use the Biomni-A1 agent scaffold and E1 environment; base and RL-trained Qwen variants are distinct models. Official model-card figure · average performance across tasks; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · CRISPR delivery · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · CRISPR delivery; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · GWAS causal gene · catalog · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · GWAS causal gene · catalog; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · GWAS causal gene · OpenTargets · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · GWAS causal gene · OpenTargets; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · GWAS causal gene · PharmaProjects · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · GWAS causal gene · PharmaProjects; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · GWAS variant prioritization · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · GWAS variant prioritization; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · LAB-Bench DbQA · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · LAB-Bench DbQA; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · LAB-Bench SeqQA · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · LAB-Bench SeqQA; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · Patient gene detection · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · Patient gene detection; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · Rare disease diagnosis · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · Rare disease diagnosis; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Biomni-Eval · Biomni-Eval1 · Screen gene retrieval · task reward · Biomni-R0 · A1 scaffold and E1 environment · 2026-10-27
Published model comparison, 2025. Biomni-R0 · A1 scaffold and E1 environment. Exact two-decimal values printed in the official task heatmap. Mean task reward is on a 0–1 scale; task-specific evaluators and sample counts differ. Only aggregate evaluation outcomes are imported. Official model-card figure · task-specific performance heatmap · Screen gene retrieval; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.

TRACE THE SCORE · Per-model notes use published breakdowns where available.

What it measures

Generalist biomedical agent evaluation suite across genetics, pharmacology, pathology, and clinical microbiology.

Reading the measurement

Ask whether the evaluation rewards recalling a fact, constructing a model, running an experiment or supporting a discovery. Those tasks require different evidence. For research claims, inspect reproducibility, uncertainty and what was actually checked.

Published evaluation

Exact labeled means across ten biomedical tasks. This is an equal-weight mean of task rewards, not necessarily micro-accuracy across all 433 instances. All six systems use the Biomni-A1 agent scaffold and E1 environment; base and RL-trained Qwen variants are distinct models.

Where this fits in an evaluation suite

A correct multiple-choice answer and a supported scientific contribution demand different evidence. Check whether the score rewards a proposal, an executed experiment, or a completed research target.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

Biomni-R0: Using RL to Hill-Climb Biomedical Reasoning Agents to Expert-Level ↗
Ryan Li, Kexin Huang, Shiyi Cao, Yuanhao Qu, Jure Leskovec · 2025

Reviewed 2026-10-27. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.

Continue exploring

All scientific discovery benchmarks →

Sources

What it measures

Reading the measurement

Published evaluation

Where this fits in an evaluation suite

A reproducible comparison

Continue exploring

Terminal-Bench-Science 0.1

SciCode

GPQA Diamond

CritPt v2