TreeFlex divides one model’s work into two roles.
Agents search more effectively by choosing which reasoning branch to revisit
TreeFlex lets one language model navigate saved reasoning states, outperforming linear and compressed-history agents on research and software tasks.
Big Tech
Dongzhuo Li · Fangzhou Cheng · Weiwei Li · Kevin Han · Haixu Ma · Zhongchang Sun · +5 more
Meta AI
Research Digest··3 min read
Li et al.
Why this paper
From Meta AI
In one line
TreeFlex, where one model picks which tree node to resume and what to do there, beats flat ReAct and FoldAgent on research and SWE tasks.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (params 4B)
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§