Platform · Product 02
Harvest
An open benchmark and dataset for evaluating models on real biological discovery tasks, so the field can measure progress on the same ground instead of on private scoreboards.
Open benchmark
A shared yardstick for discovery models.
Discovery models are easy to claim and hard to compare. Harvest fixes the tasks, the data, and the scoring so any model, ours or anyone else's, can be evaluated the same way. Open scoring means results are reproducible and arguments stay technical.
What's in the dataset
Built to reflect the tasks that actually matter for discovery.
Tasks
Real biology
Evaluation tasks drawn from genuine discovery problems, not synthetic toy settings.
Scoring
Open and fixed
Transparent metrics and a fixed protocol, so two models can be compared without ambiguity.
Access
Released openly
Made available to the research community to encourage reproducible, comparable work.
Use cases
Who Harvest is for.
Method developers benchmarking new generative or predictive models. Pharma teams deciding which tools to trust. Researchers who want an honest, reproducible comparison before committing a model to a live program.
Apply it to a program →