Solar Pro.
South Korean enterprise model with depth-up-scaling architecture optimised for retrieval workloads.
What it is.
Upstage's frontier model family. Founded by ex-NAVER Clova and Microsoft researchers including the founder. Solar Pro uses depth-up-scaling — Upstage's published technique for stacking and continuing pretraining of decoder layers — to reach frontier quality at a fraction of the training cost. Strong Korean coverage. Solar Embedding and Solar OCR extend the line into retrieval and document parsing.
Where it fits.
Korean and Japanese enterprises that need verifiable APAC data residency and Korean-native quality. Samsung, KB Financial, and SK Telecom run Solar inside internal RAG stacks. Strong fit for Korean-language insurance, banking, and legal document understanding. Solar OCR competes with AWS Textract inside Asian regional banks.
- Strongest Korean-language enterprise performance
- Depth-up-scaling reduces training cost dramatically
- On-prem path available with full governance
- Behind frontier-closed on English long-tail knowledge
- Smaller global brand awareness limits Western adoption
Frequently asked.
What is depth-up-scaling?
Upstage's published technique for building larger models by stacking and continuing pretraining of decoder layers from a smaller base. Documented in the Solar 10.7B technical report.
Who uses Solar in production?
Samsung, KB Financial, SK Telecom, and Hyundai are public references. Most deployments sit inside internal Korean-language RAG stacks with Solar Embedding and Solar OCR alongside Solar Pro.
Is Solar available outside Korea?
Yes. Upstage Console serves the API globally. The on-prem licence is available worldwide, with most non-Korean deployments inside Japanese and US enterprises.
Enterprise-focused models with strong RAG and Coral orchestration.
the 01.AI founding team's bilingual frontier model with extended context for Chinese-English enterprise.
Chinese frontier open-weights family leading coding and math benchmarks at MoE scale.