The Year in AI Papers
The year/Independent research

Paper 2601.07022

Solar Open Technical Report

Sungrae Park, Sanghoon Kim, Jungho Cho, Gyoungjin Gim, Dawoon Jung, Mikyoung Cha, Eunhae Choo, Taekgyu Hong, Minbyul Jeong, SeHwan Joo, Minsoo Khang, Eunwon Kim, et al.

Published
Jan 2026
Research lab
Independent
Citations
2
GitHub
Not linked

01 In brief

Summary

Solar Open is a 102B-parameter bilingual Mixture-of-Experts language model developed by Upstage to address data scarcity for underserved languages, focusing on Korean.

The model uses a 196,608-token BPE tokenizer, a sparse MoE architecture with 12B active parameters, and was trained on 20 trillion tokens.

Key innovations include synthesizing 4.5T tokens of high-quality, domain-specific data; a progressive curriculum that increases data quality and synthetic data ratio (up to 64%) across phases; and SnapPO, a cyclic off-policy RL framework that decouples generation, reward computation, and training for scalable multi-objective optimization.

The model achieves competitive performance on Korean benchmarks, outperforming gpt-oss-120b-high in finance (65.5 vs 62.5), law (65.5 vs 62.8), and medical (84.4 vs 75.8) domains, while maintaining strong English performance.

Training optimizations increased throughput from 4,000 to 7,200 TPS on B200 GPUs.

The methodology is presented as a blueprint for developing LLMs for other underserved languages.

02 From the paper

Abstract

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.