AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks
Open paper
Benchmark
Model
Rank
Results
agents-on-gaia-text-only-103-validation-tasks
AREX-2
#1
Accuracy: 92.2
deep-research-agents-on-browsecomp
AREX-2
#7
Accuracy: 84.0
Rank counts only results with a code link.