DeepSeek released DeepSeek-V4-Pro (1.6T total parameters, 49B active) and V4-Flash (284B total, 13B active), both Mixture-of-Experts models with MIT license and 1M token context. V4-Pro is the largest open-weights model released so far, and its pricing at $1.74/M input undercuts GPT-5.4 and Claude Sonnet 4.6 by more than half.
LocalLLaMA surfaced an MIT-licensed GigaChat 3.1 release that pairs a 702B MoE model for clusters with a 10B MoE model aimed at faster deployment and lighter inference.
A Hacker News discussion highlighted Flash-MoE, a pure C/Metal inference stack that streams Qwen3.5-397B-A17B from SSD and reaches interactive speeds on a 48GB M3 Max laptop.
NVIDIA's new Nemotron 3 Super pairs a 120B total / 12B active hybrid Mamba-Transformer MoE with a native 1M-token context window and open weights, datasets, and recipes. LocalLLaMA discussion centered on whether those openness and efficiency claims translate into realistic home-lab deployments.