Neocloud company VeselAI said on Oct. 12 it will provide global inference infrastructure for Upstage's large language model (LLM) Solar Pro 4.

VeselAI stressed the partnership as an example of a Korea-developed AI model expanding into global markets backed by infrastructure from a Korean neocloud company.

VeselAI said it focused on raising throughput while lowering per-token costs by configuring GPU resources in line with model characteristics and actual traffic patterns and by optimizing serving infrastructure.

It applied measures including improved internal cache hit rates, model quantization and faster inference technology. This raised processing speed by an average of 2 times and up to 3 times before and after adopting endpoints, the company said. It also cut per-token costs based on higher cache hit rates, achieving cost savings of up to 2 times, it added.

VeselAI said it will continue to support inference infrastructure needed for expanded use of the Solar model.

VeselAI CEO Jae-man Ahn (안재만) said, "This partnership is meaningful in that it backed global results for a model developed in Korea with Korean neocloud infrastructure." He said the company will strengthen its capabilities as an inference provider to create an environment where model providers can focus on research and product development based on accumulated inference operations experience, and to support stable adoption of enterprise open-source models.