Keep pulling the thread on Nathan Lambert.
NVIDIA has committed to using the OpenMDW license for its Nemotron models, dropping its previous custom license.
Cohere released its flagship model, Command A+, under the Apache 2.0 license, a change from the non-commercial license used for previous versions.
The open model ecosystem has become more diverse, shifting from being dominated by a few Chinese companies a year ago to including more niche companies from around the world.
Alibaba uses its open model releases as a strategy to upsell its closed models.
NVIDIA benefits from a flourishing open model ecosystem because it increases interest in and usage of its GPUs.
Product companies like JetBrains, Zed, Krea, and Photoroom train highly specialized, small models to fit their product needs, avoid reliance on closed models, and offer unique features.
The NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 model uses LatentMoE architecture for faster performance compared to similar models.
The vast majority of the training data for the Nemotron series of models is open source.
Cohere's Command A+ is a 218B-parameter Mixture-of-Experts model with 25B active parameters.
Cohere's Command A+ model can be run on a single NVIDIA B200 GPU when using 4-bit quantization.
The GLM-5.2 model is usable for everyday work and its performance is not a significant regression compared to the best available closed models.
Zyphra's current flagship model releases are a 74B-parameter MoE with 4B active parameters and an 8B-parameter MoE with 0.6B active parameters.