ROME – November 2025
Translated, a leader in AI-powered language solutions, announced that Lara, its translation AI, now supports 200 languages and introduced a new reasoning model, Lara Think. The latter improves translation quality by achieving an average human evaluation grade 40% higher than the baseline model, approaching single-pass human translation quality in the most widely spoken languages for user-generated content.
This milestone paves the way toward universal understanding, empowering enterprises and language professionals to deliver fluent, accurate translations to 99.7% of the world’s population by supporting the official languages of the countries where they live. With this release, Translated is unlocking the world’s latent demand for translation.
Launched one year ago in 11 languages, Lara now supports contextual understanding, custom instructions, adaptation, and glossaries across a broad global set of languages. By extending this capability to low-resource languages, Lara strengthens organizations' ability to operate in underrepresented yet fast-growing markets.
New languages include:
• Africa: Hausa, Shona, Swahili, Xhosa, Zulu
• India: Gujarati, Hindi, Telugu
• Asia: Azerbaijani, Georgian, Kurdish, Tagalog
Building on initial tests where Lara outperformed popular MT systems and LLMs, this update delivers further gains in performance, significantly outperforming the most recent state-of-the-art commercial systems across the most spoken languages. Internal tests show an average 25% improvement in translation quality and up to 76% for low-resource and underrepresented languages.