Trillion Labs

Another Korean model, using a DeepSeek-like architecture with MLA. Like other models from Korean labs, it focuses on English and Korean language support.

The Korean Trillion Labs is a repeated guest at the Artifacts series. This time, they are releasing a 21B reasoning model with support for English, Korean and Japanese.

A dense 70B model. However, as it was trained on only 1.5T tokens, the performance will likely not be on par with similarly-sized models outside of benchmarks. Furthermore, it comes with a rather restrictive license.