Vessel AI Runs Global Inference for Upstage's Solar Pro 4
Vessel AI processed worldwide requests for Upstage's Solar Pro 4 on GPU servers in South Korea, Europe and the US, cutting cost per token by up to half.
Vessel AI announced it operated the global inference service for Upstage's Solar Pro 4, processing user requests from around the world on its own GPU infrastructure, according to platum.kr.
The company used GPU servers in data centers located in South Korea, Europe and the United States to handle the global traffic.
Vessel AI said it coordinated launch and promotion schedules and expected demand with Upstage in order to design the processing capacity the model would need.
On the technical side, the company applied higher cache utilization and quantization techniques, raising processing speed by an average of 2x and up to 3x, while cutting cost per token by up to half.
Solar Pro 4 reached third place in usage on the global LLM routing platform OpenRouter around September.
Earlier infrastructure work
Vessel AI recently supplied GPU infrastructure for AI research and development to SK Biopharm, and signed an agreement with Neo AI to build and operate a GPU platform at the Pohang AI Data Center.
Quick answers
What did Vessel AI do for Upstage's Solar Pro 4?
It operated the global inference service, processing worldwide user requests on its GPU servers in South Korea, Europe and the United States.
How much did Vessel AI improve processing speed and cost?
It raised processing speed by an average of 2x and up to 3x, and cut cost per token by up to half, using higher cache utilization and quantization techniques.
How did Solar Pro 4 perform on OpenRouter?
Solar Pro 4 reached third place in usage on the global LLM routing platform OpenRouter around September.