Case Study: Upstage proves Solar Open’s origin and achieves full model transparency with Weights & Biases

UpStage proves Solar Open’s origin with Weights & Biases and tops benchmarks by 100% on Korean tests

Upstage, a company aiming to be a leading frontier AI lab, needed to develop a large-scale, government-backed AI model with complete transparency for all stakeholders. The challenge was not only the massive scale of the project but also the non-negotiable requirement to provide a fully auditable record of data usage, training methodology, and reproducibility to prove the model's sovereign origin. To meet this need, they utilized Weights & Biases.

By implementing Weights & Biases as their trust-and-accountability platform, Upstage tracked every aspect of their model's development. The solution allowed them to transparently demonstrate the model's entire lineage, which proved critical for verifiably refuting claims of plagiarism. The resulting model, Solar, surpassed benchmarks, achieving 100% higher performance on major Korean benchmarks than comparable models. Weights & Biases provided the structural framework that made the work credible for public-sector reporting.

UpStage