• NVIDIA and NVIDIA partner DS&G also provide remote access to B300 servers
• Lablup proposes an AX approach for enterprises and educational institutions, connecting development and performance measurement in one workflow
• Coding agent workload traces will be collected to validate serving optimization

Lablup, an AI infrastructure software company led by CEO Jeongkyu Shin, will sponsor the “2026 InThon” hackathon track hosted by Korea University College of Informatics from October 31 to November 1 and provide “Lablup Token Factory” vouchers to participating teams.

Lablup Token Factory is an LLM API provisioning service. Through its smart router, Continuum, it provides multiple models through a single endpoint and manages usage and limits by team and user. Participating students will connect the Token Factory API to coding agents to develop services and analyze data.

DS&G, an NVIDIA partner, will provide remote access to NVIDIA B300 servers during the competition. The collaboration with NVIDIA will focus on optimizing the serving of coding agent workloads. During the event, the timing and size of requests will be collected as anonymized traces to compare throughput and response latency under different serving configurations in the B300 environment. The results of the analysis of real-world coding agent traces will be published on Lablup’s technical blog.

Now in its fourth year, InThon is jointly organized by Korea University College of Informatics, the College of Informatics Student Council, and the AI-centered University initiative. Approximately 150 students majoring in software and AI will participate in either the hackathon or the datathon. Hackathon teams will develop and submit a working app or web service aligned with the given theme within 24 hours over two days and one night.

“An enterprise AX pilot is essentially a smaller version of a hackathon,” said Jeongkyu Shin, CEO of Lablup. “Multiple teams start at the same time, under the same AI environment and a fixed support limit, build different services, and then receive an independent quality assessment through judging.” He added, “Through this competition, we aim to demonstrate a structure that enables organizations to onboard motivated members quickly while maintaining financial predictability and measuring outcomes.”

Seongmin Seo, Executive Vice President at DS&G, said, “This collaboration is meaningful because it enables us to examine the performance and optimization potential of NVIDIA B300-based infrastructure based on real coding agent workloads.” He continued, “By analyzing request data generated as participants develop various AI services, we expect to assess throughput and response latency in real-world usage environments and gain relevant technical insights.” He added, “We hope this collaboration will provide a concrete opportunity to explore how high-performance GPU infrastructure can be used and to broaden understanding of AI infrastructure optimization.”

Lablup’s Token Factory concept brings together model routing, inference, and token usage and cost management. Its Continuum Router provides a single OpenAI-compatible interface for multiple LLM backends, while Continuum Hub manages token usage, policies, and costs.