大模型牌桌上，蚂蚁如何出牌？

TL;DR - Ant Group is positioning itself as the deployment layer between foundation models and real-world industries, combining its Ling 3.0 models with consumer, enterprise, healthcare, and robotics agents. Its differentiation strategy centers on embedding AI into trusted, regulated workflows rather than competing solely on model benchmarks.

• Ling 3.0 uses mixture-of-experts designs to limit inference costs: the 124B-parameter Flash model activates 5.1B parameters per token, while Tiny activates 1.3B of 7.9B.
• Consumer agents include healthcare-focused A-Fu and service-oriented A-Bao, which can execute tasks such as appointments, ordering, and payments across connected services.
• Agentar provides enterprises with agent development, evaluation, deployment, permissions, management, and roughly 200 job-specific expert templates.
• Ant’s Lingbo “general brain” extends agent capabilities to robots in pharmacies, logistics, and manufacturing, while identity, authorization, payment, auditability, and rollback systems aim to make autonomous actions trustworthy.