Infinigence AI launches Three-Body character agent with split inference

Infinigence AI and Three-Body Universe unveiled a character agent based on Liu Cixin's novels at the Three-Body Day event in Beijing on October 25. Running on the AgentWorks platform, its three-stage disaggregated inference design P-RLD-MD cuts time to first token by 51.5% and per-token cost by 37.5%.

• The agent plays the administrator of a mini-universe from the novels
• Three-stage P-RLD-MD inference is spread across separate clusters
• RadixCache prefix caching cuts inter-cluster KV data by an order of magnitude
• Time to first token falls 51.5%, per-token cost drops 37.5%