Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

ScienceAgentBench

Benchmarks

Task NameDataset NameSOTA ResultTrend
Scientific Agent Task CompletionScienceAgentBench
Success Rate (SR)43.1
40
Scientific Code GenerationScienceAgentBench (test)
SR32.8
32
Scientific DiscoveryScienceAgentBench
Score0.225
24
Defect DetectionScienceAgentBench 12 confirmed defects, 102 tasks
Recall Average (RecA)100
12
Scientific Code GenerationScienceAgentBench
SR25.5
10
Scientific Agent TaskScienceAgentBench (test)
Success Rate (SR)18.6
6
Scientific-codingScienceAgentBench deterministic-eval 2024
Pass Rate47.4
5
Showing 7 of 7 rows