GAMOW LABS / EVALUATIONS

Benchmarks

AI agents, tested on real scientific problems.
From rare disease diagnosis to the next experiment.

RAREBENCH122cases

LABBENCH20tasks

Independent evals are the best disinfectant

Connecting to results…

Enable JavaScript to load the interactive benchmark results. Read the RareBench study or the LabBench study for published findings.

01 / RARE DISEASE DIAGNOSIS

RareBench

Can an agent identify the causal findings in a rare disease case?

Read the study

Metric Find a model

Loading graded results…

All required causal findings within the first K ranked findings; K is the number of required findings for each case.

Whiskers show 95% Wilson intervals. Invalid outputs count as failures.

02 / EXPERIMENTAL REASONING

Can an agent reason from lab evidence to the right next experiment?

Read the study

MetricMean task score

Find a model

Loading graded results…

The average fraction of rubric criteria passed per task. Each task has equal weight.

Only complete, fully reviewed 20-task configurations are shown.