blink-8M tops seven static embedding models with 55.35 MTEB and 46.70 BEIR nDCG@10
tomaarsen · x · 2026-11-18
tomaarsen rounds up blink-8M's English numbers: 55.35 on MTEB(eng, v2) and 46.70 nDCG@10 on decontaminated BEIR — leading all seven static models compared at 256 dimensions. The chart shows +2.04 MTEB / +6.75 BEIR over same-size potion-base-8M, beating larger rivals like lattice-retrieval.
Mechanism: token vector lookup, averaging, normalization — no transformer forward pass; 31MB (8M) / 256MB (64M) float32 weights, MIT licensed.
More from Models
- Raschka: benchmarking new LLMs is like using IQ tests for hiring — rasbt · 2026-11-18
- Open-source Karpo-Decision-31B tops 300-model Decision Index with zero-token, one-forward-pass answers — rohanpaul_ai · 2026-11-18
- GPT-6.2 backlash is real user experience, not astroturf: dev says Astra has quirks but Sol is a solid daily driver — pvncher · 2026-11-18
- LLM protein-folding test on fresh PDB entry lands near-random structures — MoAlQuraishi · 2026-11-18
- Per-task costs span 20x: GPT-6.2 Sol at $2.63 vs Claude Fable 5.2 at $54.25 — haider1 · 2026-11-18
- Every's founder pushes back: we re-tested the model to the wire, OpenAI gave us no heads-up — kieranklaassen · 2026-11-18