Commercial Reranker Showdown: zerank-2 Leads at Every Depth, Cohere Regresses in Deep Pools
CShorten30 · x · 2026-11-16
Part 4 of the thread benchmarks three commercial cross encoders: Cohere rerank-v4.0-pro, Voyage rerank-2.5, and ZeroEntropy zerank-2.
- At k=100, the best cross encoder roughly doubles hybrid retrieval's nDCG@10 from 0.181 to 0.358.
- Deeper pools keep lifting Recall@20, but head-of-list quality saturates by k=1,000; Cohere rerank-v4.0-pro actually regresses from 0.351 nDCG@10 at k=200 to 0.313 at k=2,000.
- ZeroEntropy zerank-2 leads nDCG@10 at every depth.
This pattern echoes the Drowning in Documents paper.
Related event: Reranking Study Finds Two-Stage Pipeline Beats Fusion and Deeper Windows(13 posts)→
More from Research
- BeyondBinding challenge launches: AI-designed antibodies get wet-lab tested — QuanquanGu · 2026-11-17
- Singularity-Bench: Opus 5.5 with Claude Code beats human baselines on building AI systems — armand_ruiz · 2026-11-17
- Open Vision Banana: fully open recipe turns image generators into multi-task perception systems — NielsRogge · 2026-11-17
- WebArena-Pro debuts as largest browser-agent benchmark, top agents score 62.3% — xhluca · 2026-11-17
- NeurIPS 2026 paper: RL-trained skills transfer poorly to composed reasoning tasks — StanfordAILab · 2026-11-17
- OpenAI warns CoT monitorability will degrade as models scale, researchers push back — gleech · 2026-11-17