14-10-2025

Almawave presents the latest-generation language models in the Velvet family, to strengthen the Large Language Model landscape in Europe and ensure accessible, multilingual AI that is effective and can be adapted to a variety of use cases.

These are the new Velvet 25B – which is optimized for text processing in all of the European Union’s 24 official languages and boasts advanced reasoning capabilities and management of complex environments – and Velvet Speech 2B, a compact, versatile multimodal text-to-voice model that enables dynamic real-time interactions. Both models were designed according to a natively multilingual approach not centered around English, with the goal of preserving each European country’s linguistic and cultural characteristics and ensuring performance and inclusion for less widely spoken languages with more limited digital data.

Valeria Sandei, Chief Executive Officer of Almawave, notes that “Velvet 25B and the first Velvet Speech 2B are a continuation of the journey that has for many years seen us working on speech technologies and for more than a year on the production of European generative AI. These new models have required us to make significant investments, work unwaveringly, and collaborate with leading national and international partners. The expertise we have gained in this area means we are now capable of developing generative AI entirely within our laboratories, transforming it into concrete solutions and creating value for institutions and businesses. We also have no intention of stopping at LLMs: our vision encompasses end-applications, multimodal and multi-agent solutions that can respond to real, concrete needs in segments such as the Public Sector, Healthcare, Security, Finance, and Transportation. This is our approach in Italy, but our focus is also increasingly international.”