Demo page link. (Until July 31st)
https://open2-beta.upstage.ai
The following is an excerpt from a Digital Times article.
https://v.daum.net/v/20260723111642316
Upstage announced on the 23rd that it has released 'Sola Open 2,' its own open-weight large language model (LLM). The company also publicly released the model weights on the global open-source platform Hugging Face and published a tech report detailing the architecture and overall learning methodology.
Sola Open 2 is the latest model developed by Upstage's own technology, through the government 'Independent AI Foundation Model' project in which Upstage is one of the participating organizations. Based on its predecessor 'Sola Open,' it provides usability optimized for agents. It has a total of 250 billion parameters and applies a Mixture-of-Experts (MoE) architecture that activates only 15 billion parameters required for actual tasks. It can handle content up to a maximum of 1 million tokens and can be run with two Nvidia H200 chips when quantization is applied.
[The rest of the content is omitted]
https://x.com/hunkims/status/2079949203615453414
