Ant inclusionAI Open-Sources Trillion-Parameter MoE Base Model Ling-2.6-1T-base
Original title:inclusionAI/Ling-2.6-1T-base
AI summary
Ling-2.6-1T-base is a trillion-parameter MoE base model open-sourced by Ant inclusionAI (about 1T total parameters, 63B activated). Upgraded from Ling-2.0-1T-base, it uses a linear attention architecture mixing Lightning Attention and MLA at a 7:1 ratio, and has undergone about 9.6T tokens of transfer pre-training, continued pre-training and mid-training, with its context window expanded in stages from 4K to 256K. It surpasses its predecessor on benchmarks including MMLU (86.82), SimpleQA and LongBenchv2 (43.54). The model is intended for research only (continued pre-training, fine-tuning, distillation, etc.) and does not directly provide chat capabilities.
Key Takeaways
- 6-1T-base is a trillion-parameter MoE base model open-sourced by Ant inclusionAI (about 1T total…
- Upgraded from Ling-2.
- 0-1T-base,it uses a linear attention architecture mixing Lightning Attention and MLA at a 7:1…
- 6T tokens of transfer pre-training,continued pre-training and mid-training,with its context window…
- It surpasses its predecessor on benchmarks including MMLU (86.
Quick Facts
- Published
- Updated
- Category
- Model Releases
- AI score
- 61
Ling-2.6 用混合线性注意力把万亿 MoE 基座模型的上下文能力推到了 256K,对于研究长上下文和 MoE 的团队是个有价值的基座,但它是未对齐的预训练模型,不能直接当对话助手用。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Ant inclusionAI Open-Sources Trillion-Parameter MoE Base Model Ling-2.6-1T-base. June 2, 2026. https://new.nat6.net/en/news/cmqg2a9rm02wlslsplna4jnm6/