The authors built IterSynth around two alternating roles: a Planner that determines what information is still needed, and a Synthesizer that incorporates new evidence into an evolving summary.
Separating planning from synthesis improves long-horizon search agents
IterSynth alternates dedicated planning and evidence-synthesis roles while maintaining a compact evolving summary instead of the full search history.
Chinese Tech
Xingyu Wu · Yuchen Yan · Zhengxi Lu · Siqi Chen · Xin ZHANG · Aiting Liu · +6 more
Zhejiang University · Tencent
Research Digest··2 min read
Wu et al.
Why this paper
From Tencent and Zhejiang University
In one line
IterSynth separates planning and synthesis in deep search agents using a summary state, outperforming prior small agents.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks (3 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§