Re3Rank: 0.6B reranker with reasoning-conditioned retrieval heads beats 32B baselines
_reachsumit · x · 2026-11-04
Re3Rank is a listwise reranker built on reasoning-conditioned retrieval heads (RRHeads). A generator writes a reasoning trace over the question and candidates, and a scorer uses the trace tokens as the attention query of selected RRHeads to read relevance as continuous scores—a straight-through estimator propagates ranking loss through the discrete trace so generator and scorer train jointly. Averaged over 9 benchmarks, the 4B variant beats the strongest of 11 baseline rerankers by 3.11 NDCG@10, and even the 0.6B variant surpasses rerankers up to 32B parameters.
More from Research
- Noetra engineer's first project: building an image encoder from scratch for multimodal LLMs — HirokatuKataoka · 2026-11-04
- Japanese team trains image encoder from scratch to match SigLIP2, with better CJK performance — HirokatuKataoka · 2026-11-04
- SpecJudge author argues LLM-as-judge tools should never ask models for self-reported confidence — jokiruiz · 2026-11-04
- Tsinghua's BEMULA benchmark: no LLM agent memory system passes over 1.5% of forget requests — rohanpaul_ai · 2026-11-04
- KAIST AI's FactEmbed decouples query understanding to improve composed image retrieval — kaist-ai · 2026-11-04
- Intology's Locus auto-designs molecular models, beating GPT-6 and Claude agents — rohanpaul_ai · 2026-11-04