On April 24, 2026, Ant Group's Bailing large model team unveiled its trillion-parameter flagship model, Ling-2.6-1T, designed for immediate deployment across a wide range of tasks. This model employs a hybrid architecture that integrates MLA (Mixture-of-Layers Attention) and Linear Attention mechanisms, utilizing a 'fast thinking' strategy to enable efficient and rapid inference. It excels at generating outputs with minimal token usage, thereby significantly cutting down on both computational costs and inference time. The model's overall intelligence level is on par with GPT-5.4 (in non-reasoning mode), showcasing outstanding performance in benchmark evaluations such as AIME2026. Additionally, Ling-2.6-1T supports ultra-long contexts of up to 256K tokens and has already rolled out its API services. Ant Group also plans to open-source the model in the near future, fostering further innovation and collaboration in the AI community.
