Neobabel A Multilingual Open Tower For Visual Generation 2025 07 09
Source
cohere.com
Author
Cohere
Date
Context
NeoBabel addresses image generation for people who prompt in languages other than English. It trains on multilingual image-text material, followed by higher-resolution instruction tuning, to support English, Chinese, Dutch, French, Hindi and Persian directly. Cohere's research abstract reports state-of-the-art results on multilingual benchmarks while retaining strong English capability.
The researchers also expanded two English image-generation benchmarks into multilingual versions. That makes evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → across languages part of the contribution: an English result alone cannot establish how well a system responds to the other languages it claims to support.
Terms in this piece · Glossary
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
postCohere releases open-weights translation model North Small Translate
articleThe Data Provenance Initiative A Large Scale Audit Of Dataset Licensing And Attribution In Ai 2023 10 25
