VESSL AI powers global inference for Upstage’s Solar Pro 4

VESSL AI said it provided the global inference infrastructure behind Upstage’s Solar Pro 4, helping the model reach No. 3 usage on OpenRouter in September. The partnership highlights how Korean infrastructure can support Korean-built AI models at global scale while improving speed and lowering serving costs.

Why it matters: - VESSL AI’s work shows that inference infrastructure, not just model quality, can determine whether an AI product scales globally. - The partnership helped Solar Pro 4 handle live traffic from users around the world while cutting serving costs. - The result also points to a broader shift: Korean neocloud infrastructure is supporting Korean-developed AI models in overseas markets.

What happened: - VESSL AI said it provided global inference infrastructure for Upstage’s large language model, Solar Pro 4. - The company said it used its own optimization technology to speed up processing and reduce costs. - In September, Solar Pro 4 ranked No. 3 by usage on OpenRouter, a global LLM marketplace. - VESSL AI managed Solar Pro 4’s global inference operations during that period and handled inference requests from users worldwide.

The details: - VESSL AI and Upstage shared launch and promotion schedules and planned expected demand together before launch. - The companies designed the required inference capacity jointly. - The setup relied on GPU clusters in Korea, Europe and the US. - VESSL AI optimized its in-house inference engine for Solar Pro 4’s characteristics and real traffic patterns. - The company said those changes increased throughput and lowered cost per token. - VESSL AI improved internal cache hit rates and applied model quantization and KV cache-aware routing. - Processing was twice as fast on average and up to three times as fast as before VESSL AI’s endpoint was introduced, the company said. - Cost per token fell by up to 50% through cache-hit-rate improvements.

Between the lines: - The arrangement let Upstage focus more on model development and service expansion while VESSL AI handled capacity planning and inference operations. - The partnership suggests that inference has become a competitive layer in AI commercialization, especially for models that see sudden traffic spikes after launch. - Jaeman An, VESSL AI’s CEO, said the partnership was significant because a Korean model’s global success was powered by Korean neocloud infrastructure. - An said VESSL AI plans to strengthen its role as an inference provider so model makers can focus on research and product development while enterprises adopt open-source models more reliably.

What’s next: - VESSL AI and Upstage plan to keep working together as adoption of the Solar models grows. - VESSL AI is expanding beyond GPU provisioning into inference operations for AI models. - The company recently supplied GPU compute for AI R&D to SK Biopharmaceuticals. - VESSL AI also signed an MoU with NeoAI to build and operate a GPU cluster at the Pohang AI Data Center. - The company continues to grow its enterprise customer base and infrastructure footprint.

The bottom line: - Solar Pro 4’s OpenRouter ranking gave VESSL AI a public example of how optimized inference infrastructure can help a model scale faster, cheaper and across more regions.