1T-scale reasoning model Ring-2.6-1T released

Ring-2.6-1T is a model with a total of 1T (1 trillion) parameters, featuring a 63B active parameter structure optimized for real-world agentic workflows.

It specializes in coding agents, tool use, and long-horizon tasks, and has recorded excellent performance on major benchmarks such as PinchBench, ClawEval, TAU2-Bench, and GAIA2-search.

Key features are as follows:

• Adaptive Reasoning: Dynamically adjusts the reasoning budget between high and xhigh modes according to task complexity, reducing token overhead while maintaining performance.
• Agent Optimization: Enables efficient execution in multi-turn conversations and tool-intensive workflows.
• Efficiency: Designed with execution quality, latency, and cost-efficiency all taken into account.