Scientific discovery
DiscoveryBench-102
Scientific-programming tasks paired with an assay, compound, target, QC-result, and literature world.
About DiscoveryBench-102
What it measures
102 scientific-programming tasks: 38 deterministic and 64 visual/LLM-judge tasks.
Scope & limitations
This Blobfish suite has its own task and environment definitions. Upstream scores and reference controls are not company performance scores.
Evaluation environment
Three tasks require GPUs; visual tasks require the verifier model configured by the upstream package.