LongCat 2.0 Open Source: 1.6T Param MoE, ~48B Active
eliebakouch · x · 2026-07-05
LongCat 2.0 model weights are open-sourced under the MIT license, featuring approximately 1.6T total parameters and 48B active parameters. Compared to its predecessor, it keeps the attention shape unchanged but reduces the number of experts from 256 to 128, while adding 135B embedding parameters and an n-gram mechanism. This design choice of "reducing the number of experts" runs counter to conventional practices.
Related event: LongCat 2.0 Open-Sourced: 1.6T Parameter MoE Model(3 posts)→
More from Models
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27
- Grok Bot auto-connects to X Money accounts without opt-in, user pulls funds — noahsolomon · 2026-10-27
- Model refuses wrong psychiatric advice in deliberately misleading test — FellMentKE · 2026-10-27
- Clinical System One v0.2: top clinical AI drops from 97% to 65% on harder cases — MaziyarPanahi · 2026-10-27