2026.11.03. 06:47

Upstage, FriendliAI build translation model infrastructure based on Solar LLM

Introducing Friendli Dedicated Endpoint to optimize GPU costs

Upstage logo.
Upstage logo.

2024.07.10. The Fair Economy — AI technology company Upstage announced on the 10th that it has signed a contract with FriendliAI to build translation model infrastructure based on its large language model (LLM) 'Solar'.

According to Upstage, the company has released a Korean-English translation model, fine-tuned from the Solar LLM, in the form of an application programming interface (API). The deep learning-based translation model goes beyond simple sentence translation, inferring meaning by synthesizing context before and after a given passage and grasping the context and flow of a conversation. In evaluations using Meta's Flores benchmark, considered a barometer for machine translation evaluation, the model's accuracy surpassed that of OpenAI's GPT-4 and DeepL.

To efficiently handle translation traffic of up to about 100,000 cases per day, Upstage adopted FriendliAI's 'Friendli Dedicated Endpoint'. Through the 'auto-scaling' feature, which dynamically adjusts GPU allocation according to traffic load, the company automated service operations and significantly reduced GPU costs.

Jeon Byeong-gon, CEO of FriendliAI, said, "I am very pleased that Upstage's best generative AI model Solar and FriendliAI's best generative AI infrastructure have come together to provide companies with the best generative AI applications at the lowest cost. FriendliAI and Upstage will strive together to help companies innovate with generative AI."

Kim Seong-hun, CEO of Upstage, said, "I am glad to collaborate with FriendliAI, a leader in generative AI infrastructure, to streamline the service costs of the Solar translation model. We expect that more people will now be able to use the world's best translation model faster and more easily."

© 2026 The Fair Economy. All rights reserved.

How did you feel about this article?

Tap once to react. Tap again to undo.

0 reactions

0 Comments

No comments yet.

Latest News