Ant Bailing Releases Trillion-Level Flagship Thinking Model Ring-2.6-1T

Deep Tide TechFlow reports that on May 9, Ant Bailing officially launched the trillion-level flagship thinking model Ring-2.6-1T, specifically designed for complex real-world task scenarios. The model introduces an innovative adjustable Reasoning Effort mechanism, supporting two levels of reasoning intensity: high and xhigh. In real-world task execution evaluations, it achieved a PinchBench score of 87.6, surpassing GPT-5.4x High, Gemini-3.1-Pro high, and Claude-Opus-4.7x high. For high-difficulty reasoning tasks, it scored 77.78 on ARC-agi-V2, aligning with Gemini-3.1-Pro high and Claude-Opus-4.7x high at the same performance level. (Jin10)