Nvidia Nemotron 3 Super
https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b
Nvidia presented their new model - Nemotron 3 Super, an open hybrid Mamba-Transformer MoE model: 120B total / 12B active parameters, 1M token context. Currently available for free in Kilo Code: https://blog.kilo.ai/p/nvidia-nemotron-3-super-launch
The Hacker News post about the release gained only 13 points and 2 comments; generally, no one seems to care. Nvidia took a long time to develop this, and Qwen 3.5 has already "caught up and surpassed" many competitors.
Cursor Model Update
https://forum.cursor.com/t/introducing-composer-2/155288
https://cursor.com/blog/composer-2Composer is Cursor's own coding LLM, which delivers good results on simple tasks. Version 2 was specifically trained on long coding tasks using reinforcement learning. The model is quite affordable, with both standard and fast variants available.
Cursor Interface Update
https://forum.cursor.com/t/what-is-cursor-glass/155327
https://cursor.com/glassGlass is a completely new interface currently in early access, based on an agent command center paradigm. Some users are already complaining that the update "forcefully" installs Glass without a way to switch back yet.
https://www.youtube.com/watch?v=stRhZIrwa-w
Now agents are managed in a single space: project threads, parallel sessions, plugin marketplace, built-in browser+terminal, one-click Git, Shift+Tab planning with Mermaid diagrams, and todos.
It's a good step to stay competitive. Of course, there's a lack of original ideas, as the name sounds like an Apple interface and the look mimics the Codex app. However, a bigger problem now is that it’s no longer easy to create or open files manually. Consequently, Cursor is losing its status as an "AI IDE" where one could still write code directly (an editor for humans).
#cursor #newllmmodel