Trajectory-Level Speculation with Semantic Verification for Tool-Using Agents
Open paper
Benchmark
Model
Rank
Results
agents-on-bench-airline
SpecAgent
#1
Accuracy: 86.0
agents-on-bench-retail
SpecAgent
#1
Accuracy: 87.7
agents-on-automationbench-public
SpecAgent
#14
Success Rate: 10.5
Rank counts only results with a code link.