On February 13, Ant Group open-sourced Ring-2.5-1T, the world's first trillion-parameter reasoning model based on a hybrid linear architecture. The model achieves leading open-source performance in long-text generation, mathematical reasoning, and agent task execution, providing high-performance foundational support for handling complex tasks in the age of intelligent agents.

In terms of generation efficiency, Ring-2.5-1T reduces memory access by more than tenfold and increases generation throughput by over three times compared to the previous generation model in long-text generation scenarios exceeding 32K tokens. Regarding deep reasoning capabilities, the model scored at gold medal levels in self-assessments based on the International Mathematical Olympiad (IMO 2025) and the Chinese Mathematical Olympiad (CMO 2025), achieving scores equivalent to 35 points in IMO and 105 points in CMO.

Additionally, the model can be easily adapted to agent frameworks such as Claude Code and the personal AI assistant OpenClaw, supporting multi-step planning and tool invocation.