JD Cloud said MiniMax M3 went live today; its JoyBuilder model-development platform immediately integrated the model and opened service. Combined with JD Cloud's proprietary inference framework and optimizations — PD-separation deployment, KV cache and speculative sampling — the company said model inference throughput and response efficiency have been significantly improved.

2026-06-01

JD Cloud said MiniMax M3 went live today; its JoyBuilder model-development platform immediately integrated the model and opened service. Combined with JD Cloud's proprietary inference framework and optimizations — PD-separation deployment, KV cache and speculative sampling — the company said model inference throughput and response efficiency have been significantly improved.