Reactions from ranked influencers
5 posts
Today, we’re releasing Ling-3.0-flash—a hybrid-reasoning MoE model built for production-scale agents. 124B parameters. Just 5.1B active per token. With 1/8 of the total and 1/12 of the active parameters, it matches or beats our 1T flagship model on most benchmarks shown.
Ling-3.0 starts with native hybrid-linear attention: KDA and MLA layers stacked 5:1. KDA gives fine-grained control over long-range memory, while 1/64 expert activation makes MoE compute more efficient. It supports 256K context natively and can scale to 1M.
hybrid kda 😍
Ling-3.0 starts with native hybrid-linear attention: KDA and MLA layers stacked 5:1. KDA gives fine-grained control over long-range memory, while 1/64 expert activation makes MoE compute more efficient. It supports 256K context natively and can scale to 1M.
Ooo that's finally an impressive move from Ant Ling! This is, more or less, a Kimi K3-lite (not from Moonshot). It's more or less on par with V4-Flash, but more than 2x smaller and faster. KDA+MLA, 1/64 sparsity, very interesting
Today, we’re releasing Ling-3.0-flash—a hybrid-reasoning MoE model built for production-scale agents. 124B parameters. Just 5.1B active per token. With 1/8 of the total and 1/12 of the active parameters, it matches or beats our 1T flagship model on most benchmarks shown.
Ling-3.0-flash from @AntLingAGI looks great for production agents: 124B MoE with just 5.1B active, KDA+MLA hybrid attention, 256K context We're working with the team to bring day-0 support to SGLang, stay tuned!
Today, we’re releasing Ling-3.0-flash—a hybrid-reasoning MoE model built for production-scale agents. 124B parameters. Just 5.1B active per token. With 1/8 of the total and 1/12 of the active parameters, it matches or beats our 1T flagship model on most benchmarks shown.