Ant Ling· @AntLingAGI · X·· 2026-04-29AI 评分57
AI 导读
推文感谢Hugging Face为开源模型的持续发展提供了重要平台。文中强调，token效率是构建实用、可持续协作系统的关键。AntLingAGI最新发布的Ling-2.6-flash模型已加入这场效率竞争，该模型采用MIT许可，具备104B/7.4B激活参数，并创新性地结合了1:7 MLA与Lightning Linear混合注意力机制。其性能表现突出：推理速度高达340 tok/s，并且仅需约1500万tokens即可完成完整的Artificial Analysis评测套件。这一效率显著高于通常需要5000万至1亿tokens的前沿模型，展现出其作为高效协作伙伴的潜力。
正文 · 原文
Thanks to @huggingface for offering the place where open models can keep thriving! 🤗🤗
Token efficiency is the key to practical and sustainable co-work systems, let elephant-alpha / Ling-2.6-flash to be your best companion 😎
Token efficiency has become a major battleground, Ling-2.6-flash released by @AntLingAGI just enters the race 🔥 ✨ 104B/7.4B active - MIT license ✨ 1:7 MLA + Lightning Linear hybrid attention ✨ 340 tok/s inference (4× H20) ✨ Completes the full Artificial Analysis suite in about 15M tokens ( vs 50-100M most frontier models )在 X 查看被引用的帖子
来源：Ant Ling · x.com