LACUNA: Safe Agents as Recursive Program Holes
Open paper
Benchmark
Model
Rank
Results
deep-research-agents-on-browsecomp-plus
Lacuna (DeepSeek-V4-Flash)
#16
Accuracy: 27.1
agents-on-bench
Lacuna (deepseek-v4-flash)
#17
Accuracy: 76.0
Rank counts only results with a code link.