T-Search, an open agentic retriever for hard multi-step search
Original titleT-Search: An Open Agentic Retriever and Playground for Hard Multi-Step Search
- Outlet
- Takara TLDR
- Published
- Importance
- 3/5
Summary
Researchers present T-Search, an open-weight agentic retriever built on Qwen3.6-35B-A3B for hard multi-step search over a fixed corpus. It achieves 56.0 Recall@10 with one rollout and 61.3 with three fused rollouts across seven English and Russian benchmarks, outperforming larger open models. The team releases the model, harness, demo, and three benchmarks including TRuST, the first native-Russian hard-search benchmark.
Original excerpt
We present T-Search, an open-weight agentic retriever for hard multi-step search. Given a question and a search tool over a fixed corpus, it runs a bounded multi-round search and returns a ranked list of evidence chunks with short justifications, leaving answer generation to a downstream model, so backend and generator can be swapped without retraining. T-Search is built on Qwen3.6-35B-A3B and trained on adversarially filtered synthetic search tasks with round-sliced supervised fine-tuning followed by GSPO on a recall reward. Averaged over seven English and Russian benchmarks with gold evidence annotations, it reaches 56.0 Recall@10 with one rollout, 14.4 points above its base, and 61.3 with three fused rollouts, outperforming larger open models. We release the model, harness, live demo, and three benchmarks, including TRuST, the first native-Russian hard-search benchmark.