Yi-Fan Hu

Coordinated multi-humanoid loco-manipulation is achieved by a single decentralized policy that dynamically blends pre-trained single-humanoid skills, bypassing task-specific motion generation.

Treating every teacher token equally during on-policy distillation leaves massive gains on the table—dynamically reweighting supervision allows a smaller student model to outperform its own larger teacher on complex reasoning tasks.

Mint-Agent models not only outperform existing benchmarks in financial reasoning but also integrate long-horizon execution with auditable evidence trails, setting a new standard for financial intelligence.