Ring-2.6-1T是一款万亿参数的旗舰思维模型，专为现实世界复杂任务和生产环境构建。该模型具备可调节思维努力功能，通过动态计算机制灵活平衡认知深度、token成本和执行速度。它针对代理优化，适用于高频工作流，提供快速多步执行和工具编排，并具有SOTA稳定性。深度思维特性解锁了模型的最大能力上限，特别适合严格数学逻辑和科学研究。
蚂蚁憋了个万亿参数的大家伙，而且把可调思考和 Agent 优化当主打，明显是冲着生产级落地的，做 agent 的可以盯一下实际推理成本。
We are launching Ring-2.6-1T, a trillion-parameter flagship thinking model engineered for real-world complex tasks and production env: 🚀
- Adjustable Thinking Effort: dynamic compute mechanism to flexibly balance cognitive depth, token cost, and execution speed;
- Agent-Optimized: Built for high-frequency workflows, delivering rapid multi-step execution and tool orchestration with SOTA stability;
- Deep Thinking: Unlocks the model's maximum capability ceiling for rigorous mathematical logic and scientific research;
来源：Ant Ling · x.com