Perplexity releases open multimodal late-interaction embedding family with SOTA results
antoine_chaffin · x · 2026-10-08
Perplexity announced a new embedding model family under an open license, extending its late-interaction SOTA results. The models are multimodal (strong on both text and image benchmarks), multi-vector, and share a single embedding space enabling cross-model querying. Team members credit distillation as the key technique.
More from Models
- Astra breaks integer multiplication records in a loop, but fails all 4 ciphers — cneuralnetwork · 2026-10-08
- Anthropic's 5.5 lineup dissected: only Opus 5.5 truly special, Haiku costs 4x tokens — haider1 · 2026-10-08
- Microsoft demos local models at Windows launch, shows off GitHub Auto routing — DanWahlin · 2026-10-08
- First hands-on with GPT-6: better writing and an in-app "Intelligent UI" — mark_k · 2026-10-08
- How Are Sellers Offering $20 Claude Subscriptions for Just $5? — pizzadude_09 · 2026-10-08
- LLMs writing Blender code may beat a decade of specialized 3D generation models — CSProfKGD · 2026-10-08