Introducing Lara V2

Rome – June 4, 2025

Translated, a leading provider of AI-powered language solutions, today unveiled Lara V2, the latest evolution of its translation AI. Building on Lara’s acclaimed debut in November 2024, which professional translators consistently preferred over leading machine-translation systems in a double-blind evaluation, Lara V2 moves Translated closer to its mission of allowing everyone to understand and be understood in their own language. Human A/B testing by three expert linguists, using majority voting, showed that Lara V2’s output was preferred up to 46% more often than that of the previous model.

Now supporting 50 languages and locales, Lara V2 delivers increased accuracy, improved responsiveness, and is better equipped to meet the evolving needs of global businesses. These results confirm Lara’s ability to provide strong, consistent performance—even for companies with limited linguistic data.

The most notable improvements include:

• Glossary Integration: Lara now allows users to upload custom glossaries and translation memories to ensure more consistent terminology across all content types.
• Improved Context Handling: Lara more reliably follows user instructions for tone, style, and formatting, enabling more nuanced and accurate outputs.
• Expanded Locale Support: Lara now supports 18 additional locales, including British and American English, Canadian French, and Hong Kong Chinese, extending its reach and precision across regional language variants.
• Document translation via API: Supports Lara’s styles (faithful, fluid, and creative) and translation memories.

Lara V2 is powered by a redesigned architecture featuring a new tokenizer that addresses the issues traditional large language models face when dealing with multilingual data. It was trained on a larger, more diverse, and meticulously cleaned dataset using Trust-Attention V2, Translated’s proprietary methodology that weights high-quality data more heavily during training to produce more accurate and reliable translations. Thanks to this architecture and the newly added features, Lara V2 enables deeper integration of human translation into the localization process, achieving a level of human–AI symbiosis never seen before.

The new model is immediately available on Lara Translate’s platform and via API, with new features and language support rolling out in the coming weeks. Users of the Lara Team and Enterprise Plans, which enable localization managers to share their company’s custom-trained model with colleagues, will also have immediate access.

The Lara Translate platform has also received an upgrade. Users can now edit translations directly, either manually or using a new alternative suggestion feature, which will be available soon. Immediate feedback is provided through Lara’s pop-up, a distinctive feature that serves as a quality assurance tool.

These improvements will help individual translators, teams, and enterprises to scale their translation workflows with greater quality, consistency, and speed, making it easier for everyone to communicate clearly and effectively in their own language at any scale.