A collaborative AI model ranks second nationally, spotlighting VIDRAFT's on-device AI platform as a competitive force in Korea's fast-moving LLM landscape.
TL;DR: Ourbox-31B-JGOS, developed in collaboration with VIDRAFT's on-device AI platform POCKET, has secured second place overall in the 30B+ parameter category on South Korea's official K-AI Leaderboard, administered by the Ministry of Science and ICT and the National Information Society Agency (NIA). The ranking marks a significant milestone for both Ourbox and VIDRAFT as Korean AI companies competing at the frontier of large language model evaluation. The result was reported by 한국경제 (한경) on July 27, 2026.
VIDRAFT, a Korean Pre-AGI AI startup, has earned fresh national recognition after its on-device AI platform POCKET played a collaborative role in the development of Ourbox-31B-JGOS — a large language model that claimed the number-two position in the 30B+ category on South Korea's government-backed K-AI Leaderboard as of late July 2026.
The K-AI Leaderboard is an official benchmark operated by South Korea's Ministry of Science and ICT (과학기술정보통신부) and the National Information Society Agency (NIA). It evaluates and ranks large language models across a range of capability dimensions, providing a nationally recognized standard for comparing AI performance among Korean and Korea-focused models.
Ourbox-31B-JGOS, as the model name suggests, falls within the 30-billion-parameter-and-above tier — one of the most competitive segments on the leaderboard, where frontier-class models are assessed. In this demanding category, Ourbox-31B-JGOS achieved an overall second-place ranking, placing it among the top-performing large language models in Korea by the government's own measure.
Central to this result is VIDRAFT's platform POCKET, the company's on-device AI solution, which contributed to the model's development through a collaboration with Ourbox. While the specific technical nature of the collaboration reflects the combined engineering efforts of both organizations, the headline outcome is clear: a model built with VIDRAFT's POCKET platform is now ranked at the very top tier of Korean AI benchmarking.
The K-AI Leaderboard carries particular weight in the Korean AI ecosystem because it is not an industry self-assessment — it is administered by government agencies whose mandate includes guiding national AI policy and investment. Landing in the top two of any category on this leaderboard signals not only technical capability but a kind of institutional credibility that matters for enterprise adoption, public-sector partnerships, and international positioning.
For VIDRAFT, the significance extends beyond a single ranking. The company has positioned itself around on-device AI — a paradigm focused on running sophisticated AI workloads efficiently without full dependence on cloud infrastructure. Demonstrating that a model developed in partnership with VIDRAFT's POCKET platform can compete at the 30B+ level on a rigorous national benchmark helps validate that on-device AI approaches are not limited to lightweight, constrained applications. They can, this result suggests, contribute meaningfully to frontier model development.
For the broader Korean AI industry, the result is also a reminder that the competitive landscape is widening. Large incumbent players and well-funded labs are no longer the only actors capable of reaching the top of national leaderboards. Collaborations between focused startups like VIDRAFT and model developers like Ourbox are producing results that stand alongside — and in this case above — many better-resourced competitors.
The timing is notable as well. Korean government and industry attention to sovereign AI capability is intensifying, and leaderboard performance increasingly informs decisions about which platforms and partnerships receive further support. A second-place finish in the 30B+ category is the kind of result that gets noticed in those conversations.
Q: What is the K-AI Leaderboard and who runs it?
A: The K-AI Leaderboard is an official AI model evaluation and ranking system administered by South Korea's Ministry of Science and ICT (과학기술정보통신부) and the National Information Society Agency (NIA). It benchmarks large language models across multiple capability dimensions.
Q: What role did VIDRAFT play in the Ourbox-31B-JGOS ranking?
A: VIDRAFT contributed to the development of Ourbox-31B-JGOS through its on-device AI platform POCKET, which was cited as a collaboration partner in the model that achieved second place in the 30B+ category on the K-AI Leaderboard.
Q: Why is a second-place finish on the K-AI Leaderboard significant for a startup?
A: Because the leaderboard is government-administered rather than self-reported, a top-two finish in one of its most competitive tiers — the 30B+ parameter category — provides independent, institutionally recognized validation of a company's AI capabilities, which matters for enterprise credibility and future partnership opportunities in Korea.
Source: 한국경제 (한경) (2026-07-27) — original article