AgentBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
AgentBench OS
37.6Accuracy
15
Lifelong AgentBench
100Accuracy
14
AgentBench FHIR (val)
76.7Score
8
AgentBench
34.1Success Rate
8
AgentBench
0.77Pearson r
8
AgentBench topology simulation seed=42 (n=50 per topology)
1.39ρ(B)-reduction
7
AgentBench Web Shopping
36.2Task Completion Score (TCS)
4
AgentBench WS
52Task Completion Score (TCS)
4
AgentBench LTP
32.3Task Completion Score (TCS)
4
AgentBench DCG
73.6Task Completion Score (TCS)
4
AgentBench KG
72.7Task Completion Score (TCS)
4
AgentBench DB
74.3Task Completion Score (TCS)
4
AgentBench 700 (test)
79.7Success Rate
3
AgentBench FB15k-237
83.3Success Rate (SR)
2
AgentBench DB-Bench 20 held-out tasks
16.7CheckBeforeSubmit (CBS)
1