Ant group just dropped Ring-2.6-1T ๐ฅ 1T reasoning model, built for real world agent workflows.
โจ MIT license
โจ 128K >> 256K context (YaRN)
โจ Async RL + IcePop training architecture
โจ Dual reasoning effort: "high" for fast agent loops, "xhigh" for deep reasoning = Better cost/performance tradeoff ๐