Embeddings & Vector SearchLiang Wang et al. (Microsoft)Microsoft

Microsoft researchers present E5

Presented E5 — contrastive training on large-scale weakly supervised pair data; claimed to be the first model to beat BM25 zero-shot on BEIR

Why it matters

A rebuttal to the challenge BEIR had raised. The starting point of the open-weight embedding race

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2212.03533

Added in Week 35, 2026 · See that week

Same week

Task ArithmeticPaperEU Council adopts its general approach on the AI ActPress

Nearby on the same axis

OpenAI releases text-embedding-ada-002OfficialHugging Face and Cohere release MTEBPaperGoogle DeepMind researchers present SigLIPPaperChroma announces $18 million seed roundOfficial