New AI Model by Unbabel Outperforms OpenAI’s GPT-4 in Language Translation

Unbabel, a tech company specializing in translation services, has developed a new AI model called TowerLLM that surpasses OpenAI’s GPT-4 and other AI systems in translating between English and six commonly spoken European and Asian languages. The competition in the field of language translation has been fierce, with large language models like GPT-4o dominating the market.

In a series of tests, Unbabel compared TowerLLM against GPT-4o, GPT-4, GPT-3.5, and models from Google and DeepL. TowerLLM showed superior accuracy in translating English to Spanish, French, German, Portuguese, Italian, and Korean, outperforming its competitors by a small margin. Additionally, the model excelled in translating documents for specific professional domains such as finance, medicine, law, and technical writing.

According to João Graça, Unbabel’s chief technology officer, TowerLLM’s success can be attributed to its multilingual training on a diverse dataset and fine-tuning with high-quality translations. The model’s performance highlights the significance of data quality in training AI models and suggests that smaller models can achieve remarkable results when trained effectively.

TowerLLM is currently available in two sizes with different parameters and supports 18 language pairs. Unbabel plans to expand the model’s language support and enhance its capabilities for specific business translation tasks in the future.

For more updates on AI advancements in business, subscribe to the Eye on AI newsletter for free.