Event ·Multi-source

poolsideai 发布 Laguna S 2.1：118B MoE 开源模型，8B 激活

poolsideai 发布了最新开源模型 Laguna S 2.1，总参数量 118B，采用 MoE 架构，每个 token 仅激活 8B 参数。该模型支持 1M token 上下文窗口，并提供思考与免思考两种模式。权重已在 Hugging Face 以 OpenMDW-1.1 许可证开放，可在单台 NVIDIA DGX Spark 上本地运行。Laguna S 2.1 在性能上可与参数量大数倍的模型竞争。

Current read

poolside 新出的 Laguna S 2.1 开源又高效，118B 模型只激活 8B 就能跑，还支持 1M 上下文，值得一试。

11 sources67° AI heatLast updated 2026/7/21 17:08:07

Evidence

11 sources

Primary sourceNVIDIA AI

2026/7/21 17:08:07View source

Related source 1elvis

2026/7/21 22:58:44View source

Related source 2marktechpost

2026/7/22 00:01:32View source

Related source 3vLLM

2026/7/23 07:13:47View source

Related source 4LMSYS Org (SGLang)

2026/7/23 14:37:47View source

Related source 5Decoder

2026/7/23 12:24:53View source

Related source 6Thomas Wolf

2026/7/22 08:46:49View source

Related source 7Latent Space (swyx)

2026/7/23 05:09:14View source

Related source 8IT之家

2026/7/20 03:57:36View source

Related source 9Aravind Srinivas

2026/7/22 01:54:12View source

Related source 10AWS Machine Learning Blog

2026/7/20 17:01:32View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story