_version_ 1866912817340219392
author Park, Sungrae
Kim, Sanghoon
Cho, Jungho
Gim, Gyoungjin
Jung, Dawoon
Cha, Mikyoung
Choo, Eunhae
Hong, Taekgyu
Jeong, Minbyul
Joo, SeHwan
Khang, Minsoo
Kim, Eunwon
Kim, Minjeong
Kim, Sujeong
Kim, Yunsu
Lee, Hyeonju
Lee, Seunghyun
Lee, Sukyung
Park, Siyoung
Shin, Gyungin
Song, Inseo
Song, Wonho
Yang, Seonghoon
Yi, Seungyoun
Yoon, Sanghoon
Ko, Jeonghyun
Song, Seyoung
Choi, Keunwoo
Lee, Hwalsuk
Kim, Sunghun
Chang, Du-Seong
Cho, Kyunghyun
Choe, Junsuk
Lee, Hwaran
Lee, Jae-Gil
Lim, KyungTae
Oh, Alice
author_facet Park, Sungrae
Kim, Sanghoon
Cho, Jungho
Gim, Gyoungjin
Jung, Dawoon
Cha, Mikyoung
Choo, Eunhae
Hong, Taekgyu
Jeong, Minbyul
Joo, SeHwan
Khang, Minsoo
Kim, Eunwon
Kim, Minjeong
Kim, Sujeong
Kim, Yunsu
Lee, Hyeonju
Lee, Seunghyun
Lee, Sukyung
Park, Siyoung
Shin, Gyungin
Song, Inseo
Song, Wonho
Yang, Seonghoon
Yi, Seungyoun
Yoon, Sanghoon
Ko, Jeonghyun
Song, Seyoung
Choi, Keunwoo
Lee, Hwalsuk
Kim, Sunghun
Chang, Du-Seong
Cho, Kyunghyun
Choe, Junsuk
Lee, Hwaran
Lee, Jae-Gil
Lim, KyungTae
Oh, Alice
contents We introduce Solar Open, a 102B-parameter bilingual Mixture-of-Experts language model for underserved languages. Solar Open demonstrates a systematic methodology for building competitive LLMs by addressing three interconnected challenges. First, to train effectively despite data scarcity for underserved languages, we synthesize 4.5T tokens of high-quality, domain-specific, and RL-oriented data. Second, we coordinate this data through a progressive curriculum jointly optimizing composition, quality thresholds, and domain coverage across 20 trillion tokens. Third, to enable reasoning capabilities through scalable RL, we apply our proposed framework SnapPO for efficient optimization. Across benchmarks in English and Korean, Solar Open achieves competitive performance, demonstrating the effectiveness of this methodology for underserved language AI development.
format Preprint
id arxiv_https___arxiv_org_abs_2601_07022
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Solar Open Technical Report
Park, Sungrae
Kim, Sanghoon
Cho, Jungho
Gim, Gyoungjin
Jung, Dawoon
Cha, Mikyoung
Choo, Eunhae
Hong, Taekgyu
Jeong, Minbyul
Joo, SeHwan
Khang, Minsoo
Kim, Eunwon
Kim, Minjeong
Kim, Sujeong
Kim, Yunsu
Lee, Hyeonju
Lee, Seunghyun
Lee, Sukyung
Park, Siyoung
Shin, Gyungin
Song, Inseo
Song, Wonho
Yang, Seonghoon
Yi, Seungyoun
Yoon, Sanghoon
Ko, Jeonghyun
Song, Seyoung
Choi, Keunwoo
Lee, Hwalsuk
Kim, Sunghun
Chang, Du-Seong
Cho, Kyunghyun
Choe, Junsuk
Lee, Hwaran
Lee, Jae-Gil
Lim, KyungTae
Oh, Alice
Computation and Language
We introduce Solar Open, a 102B-parameter bilingual Mixture-of-Experts language model for underserved languages. Solar Open demonstrates a systematic methodology for building competitive LLMs by addressing three interconnected challenges. First, to train effectively despite data scarcity for underserved languages, we synthesize 4.5T tokens of high-quality, domain-specific, and RL-oriented data. Second, we coordinate this data through a progressive curriculum jointly optimizing composition, quality thresholds, and domain coverage across 20 trillion tokens. Third, to enable reasoning capabilities through scalable RL, we apply our proposed framework SnapPO for efficient optimization. Across benchmarks in English and Korean, Solar Open achieves competitive performance, demonstrating the effectiveness of this methodology for underserved language AI development.
title Solar Open Technical Report
topic Computation and Language
url https://arxiv.org/abs/2601.07022