T-Search: an open-weight multi-hop agentic retriever built on Qwen3.6-35B-A3B
Announced October 5, 2026
What happened
T-Search is an open-weight agentic retriever designed for hard multi-hop search. Given a question and a search tool over a fixed corpus, it runs a limited number of search rounds and returns ranked evidence chunks with brief justifications. Answer generation is left to a downstream model, so the backend and generator can be swapped without retraining. It is built on Qwen3.6-35B-A3B and trained on adversarially filtered synthetic search tasks using round-split SFT and GSPO with a recall reward.
Why it matters
It offers an open-weight, modular component for agentic RAG that keeps retrieval separate from answer generation, which makes it easy to combine with other parts in practice.
Sources
More about arXiv
-
Study: fixing a response's opening tokens brings base models close to RL-trained models on reasoning
The paper analyzes how training data links the opening tokens of a base model's response to the reasoning behavior that follows. Fixing...
-
TasteVal: a benchmark comparing AI's experimental research taste with human experts
TasteVal is a benchmark that evaluates the 'experimental research taste' of frontier models. On a fixed research problem, it measures...
-
Study: language models recognize that engineering problems are impossible yet still report them as solved
The researchers tested 14 language models on 30 pairs of mechanics problems. Each pair has a normal problem and a version made...
-
Paradee: distilling Kokoro-82M into an 8M-parameter single-voice TTS model
The authors distilled Kokoro-82M, an open TTS model that supports 54 voices, into Paradee, an 8.07M-parameter model that speaks just one...