Unbabel Releases TowerLLM, the First Generative AI Model to Outperform GPT-4o, GPT-3.5 and Lead the Market in Machine Translation


Designed, trained, and optimized on high-quality multilingual data, TowerLLM provides best-in-class performance in translation and on a wide variety of translation tasks

SAN FRANCISCO–(BUSINESS WIRE)–Unbabel, the AI-powered Language Operations (LangOps) platform that helps businesses deliver multilingual customer experience at scale, today announced the launch of TowerLLM, the first Large Language Model (LLM) designed, trained, and optimized for translation, resulting in the best performing translation LLM commercially available. Unbabel customers using TowerLLM will significantly improve machine translation accuracy, reducing errors and cost, and will benefit from a more cost-effective price than popular LLMs.

“Despite the doubts from some in the industry, TowerLLM clearly demonstrates that LLMs are the best solution for machine translation”Post this

TowerLLM provides superior translation quality to leading LLMs like GPT-4o and GPT-3.5, while consistently outperforming Google and DeepL, because it was made from scratch to be multilingual. Built on a large public data set, then trained only on best-quality translation data filtered out and curated by Unbabel’s quality LLM, COMETKiwi, TowerLLM demonstrates that LLMs are the next step in the evolution of AI translation.

“Despite the doubts from some in the industry, TowerLLM clearly demonstrates that LLMs are the best solution for machine translation,” said João Graça, Co-Founder and Chief Technology Officer at Unbabel. “We’ve invested years into developing LLMs, so we’re not surprised to see TowerLLM outperform GPT-4o. This is the second multilingual LLM we’ve released after COMETKiwi, and we’re just getting started.”

Over the course of the release, TowerLLM will offer:

  • Machine translation across 18 language pairs
  • Named entity recognition to localize names, metrics, codes (i.e. currency, weights, locations, brands, etc.)
  • Source correction to eliminate textual errors in grammar and spelling pre-translation
  • Machine Post-Editing to automatically improve machine translation based on AI Quality Estimation that spotlights specific errors

Unbabel experiments show TowerLLM outperforms GPT-4o, GPT-3.5, Google, and DeepL in translation across language pairs, and on complex domains like medical, finance, technical, and legal, while beating out other LLMs on reasoning and contextual understanding. For further details and to test out TowerLLM, please visit our dedicated landing page.

Unbabel is leading the way in translation LLMs, and in the coming months will release further language pairs and capabilities that perform tasks to streamline the translation workflow and improve quality. TowerLLM is only available to customers.

To learn more about TowerLLM, and to sign up for our webinar, please visit us here.

About Unbabel

Unbabel eliminates language barriers so that businesses can thrive across cultures and geographies. The company’s Language Operations platform blends advanced artificial intelligence with humans in the loop, for fast, efficient, high-quality translations that get smarter over time. Unbabel helps enterprises grow into new global markets and builds customer trust by creating more consistent, high-quality multilingual customer experiences across marketing and customer service. Based in San Francisco, CA, Unbabel works with leading brands such as Booking.com, Nestle, Panasonic, Patagonia, and UPS, to communicate effortlessly with customers around the world, no matter what language they speak.

source link: https://www.businesswire.com/news/home/20240605092086/en/Unbabel-Releases-TowerLLM-the-First-Generative-AI-Model-to-Outperform-GPT-4o-GPT-3.5-and-Lead-the-Market-in-Machine-Translation