Ant Group BaiLing releases Ling & Ring 2.6 tech report, all three models open-sourced

蚂蚁百灵发布 Ling & Ring 2.6 技术报告

Ant Group BaiLing published full architecture, pretraining, post-training, and agent RL details for Ling-2.6-flash, Ling-2.6-1T, and Ring-2.6-1T. All three use a Hybrid Linear Attention that mixes Lightning Attention and MLA at a 7:1 ratio. Ling-2.6-flash hits 340 tokens/s decoding on 4×H20 hardware. Ling-2.6-1T shows roughly 4× token efficiency gain over its predecessor on the Artificial Analysis Intelligence Index. Ring-2.6-1T high scores 87.60 on PinchBench and 63.82 on ClawEval. Code and weights are open.

Why it matters: Ant Group's BaiLing team open-sourced three models with a Hybrid Linear Attention design blending Lightning Attention and MLA at 7:1, backed by concrete long-context efficiency data. Code and weights are public, making this a verifiable release. Not scoring higher because Ant'...