Additionally on February 13, Ant Group introduced the open-source launch of Ring-2.5-1T, the world's first trillion-parameter reasoning mannequin constructed on a hybrid linear structure, attaining open-source main efficiency in long-text era, mathematical reasoning, and agent job execution .
Mannequin weights and inference code have been printed on main open platforms together with Hugging Face and ModelScope . The corporate positions the mannequin as foundational infrastructure for dealing with complicated duties within the rising AI agent period.
In long-text era duties exceeding 32,000 tokens, Ring-2.5-1T exhibits important effectivity positive aspects. In contrast with its earlier era, reminiscence entry necessities have been diminished by greater than tenfold, whereas era throughput has elevated greater than threefold .
In inside testing, the mannequin reportedly achieved gold-medal-level scores in simulations of the Worldwide Mathematical Olympiad (IMO 2025) and the Chinese language Mathematical Olympiad (CMO 2025), scoring 35 and 105 factors respectively .
Ring-2.5-1T is predicated on the Ling 2.5 structure. By way of optimized consideration mechanisms, the mannequin's activation parameters have elevated from 51B within the earlier era to 63B . The mannequin is appropriate with agent frameworks resembling Claude Code and the OpenClaw private AI assistant, supporting multi-step planning and gear utilization .
Supply: CLS AIdaily
Elevate your perspective with NextTech Information, the place innovation meets perception.
Uncover the most recent breakthroughs, get unique updates, and join with a world community of future-focused thinkers.
Unlock tomorrow’s traits as we speak: learn extra, subscribe to our publication, and turn out to be a part of the NextTech neighborhood at NextTech-news.com
