NVIDIA's new multimodal reranker lifts ViDoRe v3 score from 58.82 to 67.56
tomaarsen · x · 2026-10-22
tomaarsen highlights two NVIDIA models announced at the GTC keynote: a smaller embedding model and a multimodal reranker for teams searching mixed image+text document collections.
- On ViDoRe v3, reranking lifts Embed 1B's result from 58.82 to 67.56 nDCG@10, beating Qwen3-VL-Reranker-8B's 66.03.
- Both rerankers score the same top 50 candidates from Embed 1B, making this a fair comparison of the second retrieval stage.
Related event: NVIDIA Open-Sources Nemotron 3.5 Retriever Suite: Embed, Rerank, Parse(12 posts)→
More from Models
- Grok Bot now live in X Chat, handling calendars and inbox — elonmusk · 2026-10-23
- AI2's OlmoEarth embeddings beat AlphaEarth and Tessera v2 on 10 of 11 evals — allen_ai · 2026-10-22
- Developer attempts to burn through a week of Codex usage in just 4 hours — jasonkneen · 2026-10-22
- Testing decision models like Jev and Drex Max for research triage: Drex Max heads to production — omarsar0 · 2026-10-22
- LlamaIndex releases OCRia-1, a 4B document-parsing VLM claiming SOTA on ParseBench at 87.4 — llama_index · 2026-10-22
- After MoE and MTP, local dense models haven't seen architectural speedups — Express_Quail_1493 · 2026-10-22