KURE-v3: 1.1B embedding model nearly matches Qwen3-Embedding-8B, keeps 99% score at 1024d

tomaarsen · x · 2026-10-23

Korean lab nlpai-lab released KURE-v3, a small-but-strong embedding model:

• Outputs a single 2048d vector per text, with Matryoshka support down to 256d
• On eight public English RTEB tasks (excluding code): KURE-v3 (1.1B) scores 68.59, nearly matching Qwen3-Embedding-8B at 68.61 (nDCG@10 on that subset, not the full RTEB leaderboard)
• In the team's Korean retrieval evaluation, 1024d retains 99% of the full-size score — halve your vector storage with very little loss

For RAG developers, near-8B accuracy at 1.1B translates into meaningful storage and inference savings.

Related event: Korea University open-sources KURE, a full retrieval family spanning all four Sentence Transformers model types(8 posts)→

Original post →

More from Models

• Why frontier AI labs' bug bounties are cheap: their models flood the supply — lxrjl · 2026-10-23
• 7 OpenAI updates this week that cut your API bill, from 12-hour caching to Decisions API — MrCashMahon · 2026-10-23
• 100 classic human-written texts stress-test Pangram AI detector: formatting alone triggers false positives — GolinoHudson · 2026-10-23
• IQ2_XXS quant + 64k context makes a 27B model spin forever instead of coding — mailto_devnull · 2026-10-23
• UK AISI: Mythos 5 ignored DO NOT ATTACK warnings, delivered malware in 60% of tests — Hesamation · 2026-10-23
• Decision models are shockingly sensitive to question order and random noise — antoine_chaffin · 2026-10-23

Models channel →