VultronRetriever Model Family Released
madkimchi · reddit · 2026-07-11
The VultronRetriever series of retrieval models has been released on Hugging Face, emphasizing fully offline Q&A and document embedding on iPhones. The series achieved strong scores on MTEB: Prime-8B ranked first in its category and claims the overall #1 spot; Core-4.5B ranked second; Flash-0.8B also runs on edge devices with support for higher throughput.
The author also highlighted several engineering selling points:
- Prime-8B's index storage footprint is up to 16x smaller than the previous generation of 9B-class leading models, with 12x higher throughput;
- Flash-0.8B works fully offline with an indexing speed of 60 images per minute;
- Uses Hydra Architecture to perform late interaction retrieval + generation with lower VRAM;
- Training claims 0% cross-dataset duplication and 0% eval contamination.
More from Models
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26
- AI search APIs ranked: top 12 within 7 points, yet prices differ nearly 7x — FinanceYF5 · 2026-10-26
- OpenAI Dots coming to EU/EEA and UK in a few weeks, slides confirm — testingcatalog · 2026-10-26
- OpenAI Dots demoed live as EU launch speculated, Codex adoption in Germany revealed — testingcatalog · 2026-10-26
- "Abliterated models just have brain damage" — community pushes back on unaligned model hype — dhadfieldmenell · 2026-10-26