RenderRank reranks text as images, using 16-35% fewer tokens with better NDCG
_reachsumit · x · 2026-09-29
RenderRank renders document text as images and scores relevance from compressed visual tokens via a vision-language model, cutting input length—a compounding win in reranking where every candidate is scored. Training first aligns visual relevance scores with a text teacher, then refines relative positive/negative scores. On 11 BEIR datasets it uses 16.5-35.5% fewer tokens with average NDCG@10 of 55.96, beating all sub-4B text baselines; on four long-document sets it hits 88.27 NDCG@10 with half the input tokens and 1.70x the best baseline throughput.
More from Research
- NYU talk argues 30+ years of consciousness research may be partly wrong — birchlse · 2026-10-24
- Psycholinguistic analysis finds Claude Opus 5 the most out-of-distribution model tested — gleech · 2026-10-24
- Google Research tests 180 agent setups: multi-agent hurts sequential tasks by 39-70% but boosts parallel ones 80.9% — ombre-ai · 2026-10-24
- Loopholing: pushing self-conditioning all the way in discrete diffusion language models — NandoDF · 2026-10-24
- Review-it tool suggests 10 best-fit journals and conferences for your paper — Faheem_uh · 2026-10-24
- As AI Generates Thousands of Scientific Hypotheses, Verifying Machine Reasoning Calls Physics Back — AryHHAry · 2026-10-24