“MoonMath delivered meaningful performance gains for our image inference workloads. Their low-level optimizations translated directly into faster generation and better GPU efficiency.”

“We partnered with MoonMath to optimize our BADAS world model. Within a few weeks, they introduced a novel algorithm with promising potential for significant latency reduction.”

“For LTX-2, MoonMath demonstrated a communication solution that significantly outperformed NCCL, showing the kind of low-level GPU expertise that can materially improve inference performance.”

“MoonMath impressed us with their unique ability to identify the high-level bottlenecks of diffusion transformer systems, develop intuition for promising areas of optimization, and bring the technical depth in low-level AI performance needed to materialize these improvements across major GPU vendors. Their work sits exactly at the intersection of kernels, systems, and model efficiency that matters for scaling diffusion and world-model inference.”