Reproduce “Read, Route, Patch: Robust Test-Time Evolution for LLM Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kaijie Chen, Liang Yan, Yilan Wei, Jiachen Luo, Zhihao Dou, Jiashu Qu, Haoran Liu, Junli Wang
Trident extends EvoTest’s frozen-backbone, gradient-free Act–Evolve loop for long-horizon agents by selecting useful memories, routing among behaviorally diverse harnesses, and applying local, ordered updates. On qwen-32B, it scores 0.4051 versus APEX’s 0.3100 across five benchmarks, a reported 30.7% relative gain on a mixed-metric macro-average; it ranks first in 13 of 15 settings, not all.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.