Task library

A preview of specific tasks we use to evaluate agents' biological reasoning abilities

17Tasks in the collection
17 tasksScores use each task's own rubric
LabBench15 criteria

Assess a hidden binding pocket at the NRF2–BTRCP interface

Structural biologyProtein interactionsAgent graded
GPT 6.1 sol46.7%
LabBench15 criteria

Assess the cell-type information carried by non-promoter transcription signatures

TranscriptomicsGene regulationAgent graded
GPT 6 astra40.0%
Sequencing analysis2 criteria

Assess ER folding regulation in the MCF7 treatment experiment

TranscriptomicsProtein foldingDeterministic
GPT 6.1 sol0.0%
Sequencing analysis3 criteria

Choose a microRNA source study and test its claim

TranscriptomicsNoncoding RNADeterministic
GPT 6.1 sol33.3%
Sequencing analysis9 criteria

Resolve the dose dependence of AP-1 transcription

TranscriptomicsGene regulationDeterministic
GPT 6 astra44.4%
Sequencing analysis4 criteria

Resolve the glycolytic transcription response

TranscriptomicsMetabolismDeterministic
GPT 6 astra50.0%
Rare disease genomics10 criteria

SLC25A46 compound heterozygote with an intronic deletion

Human geneticsMitochondrial biologyAgent graded
fable 5 153.4%
Rare disease genomics10 criteria

COL6A1 deep-intronic variant in suspected muscular dystrophy

Human geneticsRNA splicingAgent graded
fable 5 116.1%
deepseek v4 pro98.9%
Rare disease genomics10 criteria

RNU2-2 variants in developmental and epileptic encephalopathy

Human geneticsNeurogeneticsAgent graded
fable 5 118.2%
deepseek v4 pro25.0%
Rare disease genomics10 criteria

DYSF frameshift with a deep-intronic pseudoexon variant

Human geneticsRNA splicingAgent graded
fable 5 180.0%
deepseek v4 pro61.3%
Literature workflows6 criteria

Assess repeatable changes in A549 cell morphology

Cell morphologyPharmacologyDeterministicAgent graded
GPT 6 astra0.0%
Literature workflows10 criteria

Resolve the selective early gene response to neuroactive drugs

TranscriptomicsCancer biologyDeterministicAgent graded
GPT 6 astra10.0%
Literature workflows7 criteria

Separate potent non-oncology activity from broad high-dose killing

Cancer biologyPharmacologyDeterministicAgent graded
GPT 6 astra28.6%
Literature workflows7 criteria

Can RNA identify disrupted mitochondrial machinery in an unseen target?

TranscriptomicsMitochondrial biologyDeterministicAgent graded
GPT 6 astra42.9%
LabRobotBench29 criteria

Finishing a time-sensitive dose-response assay

PharmacologyCell viabilityDeterministic
GPT 6.1 sol32.2%
opus 5 532.1%
gemini 4 argon32.1%
LabRobotBench43 criteria

Finishing an interrupted cell assay

Cell biologyCell viabilityDeterministic
GPT 6.1 sol48.6%
opus 5 55.1%
gemini 4 argon17.1%
LabRobotBench35 criteria

Rescuing assays and cold stocks

Cell biologyBiochemical assaysDeterministic
GPT 6.1 sol59.4%
opus 5 531.7%
gemini 4 argon13.4%