AI Briefing#moe
The latest AI and developer news about #moe, with the original source and a short summary.
The latest AI and developer news about #moe, with the original source and a short summary.
A look at efficient training methodologies and the latest technology trends for foundation models, which serve as the basis of LLMs.
An inference engine written in C has been released that keeps a large MoE model's weights on disk and streams only the parts needed, allowing it to run even on low-spec PCs.
NVIDIA has unveiled Molt, a PyTorch-native RL framework designed for agentic training of large-scale MoE models.
MTP-based speculative decoding and support for Hy3, Tencent's 299B MoE model, have been added to llama.cpp.
Moondream 3.1, a model that applies MoE architecture to boost efficiency and performance, has been released.
How to maximize performance by optimizing GLM-5.2 with NVFP4 and TP=4 settings in an NVIDIA B200 environment
NVIDIA has released Nemotron-3 Puzzle-75B, an MoE-based model that dramatically improves inference efficiency.
NVIDIA has released Audex-30B, a unified multimodal MoE-based model that processes text and audio simultaneously.
Meituan has released LongCat-2.0, an open-source MoE model with 1.6T parameters trained on AI ASICs.
This article introduces engineering optimization techniques that leverage the Hybrid SWA architecture of the MiMo-V2.5 model to reduce KVCache storage space and compute to about 1/7 the level.
Tencent has released Hy3, a 295B-parameter MoE model, under the Apache 2.0 license.
ELDR routing, which accounts for expert locality in MoE models, cuts decode latency by up to 13.9%.
Meituan has released LongCat-2.0, a 1.6-trillion-parameter MoE model, in API form, strengthening agentic coding and long-context processing capabilities.
Huawei has open-sourced the OpenPangu-2.0-Flash model, which supports a 512K context length.
LongCat-2.0, a large-scale MoE-structured model with a total of 1.6 trillion parameters, has been released.
FuriosaAI has dramatically accelerated model support speed with SDK 2026.3, which introduces the new kernel framework TCL.
Minimax has open-sourced the weights of the M3 MoE model, with 428B total parameters and 23B active parameters.
Moonshot AI has released Kimi K2.7 Code, a model that significantly enhances coding task efficiency and agentic capabilities.
Zyphra has launched ZONOS2, an open-source MoE-based TTS model capable of high-fidelity voice cloning and real-time inference.
Cohere has released North Mini Code, an open-source agentic coding model for developers.
Luce Spark has been released, learning real-time routing patterns to efficiently keep MoE experts resident on the GPU.
They unveiled **InstinctRazor**, a technology optimized to run MoE models on edge devices.
InstinctRazor technology has been unveiled, running a 122B MoE model even in an 8GB VRAM environment by placing experts on the CPU.
JetBrains has unveiled **Mellum 2**, a 12B-parameter MoE model specialized for software engineering.