1T
128K -> 256K (YaRN)
MIT
Official open-weight trillion-parameter Ring reasoning flagship aimed at real-world complex tasks, especially agent workflows, engineering development, scientific research analysis, complex business systems, and enterprise automation. The official model card positions Ring-2.6-1T as an execution-first model with stronger multi-step task decomposition, tool use, contextual planning, error correction, and long-horizon workflow stability than prior releases. InclusionAI highlights two adjustable reasoning-effort modes, high and xhigh, so developers can trade off speed, token overhead, and reasoning depth depending on task difficulty. The release also emphasizes an asynchronous reinforcement learning training pipeline combined with the IcePop algorithm to improve training efficiency and stability for trillion-parameter long-horizon RL. Official benchmark claims include 87.60 on PinchBench, 63.82 on ClawEval, and 95.32 on Tau2-Bench Telecom for agent execution, plus 66.18 on ARC-AGI-V2, 95.83 on AIME 26, and 88.27 on GPQA Diamond in xhigh reasoning mode. The official download table lists 128K context with expansion to 256K via YaRN, and the model card includes direct deployment guidance for Transformers, vLLM, Docker, and SGLang.