23 results for “MoE”
NVIDIA released Nemotron 3.5 Lightning, a 30B MoE model with 3B active parameters built for high-volume agent execution. The open model cuts inference cost for always-on agents by handling routine tool calls while frontier models focus on planning.
Tencent's Hunyuan team released Hy-MT2 on 21 May 2026 — a new translation model family with instruction-following templates, three sizes up to a 30B-A3B MoE variant, and open weights plus IFMTBench. This is the next generation after the April Hy-MT1.5 phone quantizations.
Tencent open-sourced an extreme quantized GGUF build of its 770B Hunyuan Hy4 model, compressing 1.5TB weights down to 214GB. By varying precision from 1.31-bit on non-critical layers to 2+ bits on sensitive layers, the model fits on accessible multi-GPU nodes.
Alibaba's Qwen team released Qwen3.8-Flash-Next, a 125B-parameter multimodal MoE model with 6B activated parameters and 51B N-gram embeddings. The open-weight preview of Qwen4's architecture cuts training cost to one-ninth of its predecessor while leading coding and office-task benchmarks.
DeepSeek V4 Flash is positioned as a fast, lower-cost open-weight AI model for developers who need long-context reasoning without paying frontier-model. The model is described as a Mixture-of-Experts system with 284 billion total.
NVIDIA validated Alibaba's Qwen3.8-Flash-Next 176B MoE model on its GB300 NVL72 rack, achieving over 16K tokens per second per GPU. The hybrid GDN/QSA architecture keeps memory and compute bounded for million-token agentic coding workloads.
NVIDIA published Day-0 FP8 inference results for Alibaba's 2.4T-parameter Qwen3.8-2.4T-A95B on the GB300 NVL72 rack, hitting over 4,000 tokens per second per GPU. The result demonstrates that rack-scale NVLink can absorb MoE all-to-all traffic at frontier model scale.
Tencent open-sourced Hy4 preview, a 770B-parameter MoE model with 49B activated weights and a 1M-token context window. The Apache 2.0 release puts a frontier-class model within reach of self-hosted inference stacks.
AMD, Super Micro Computer and Spectro Cloud have jointly unveiled AMD Instinct Coder, a turnkey enterprise AI coding platform designed to give. Enterprise AI coding has moved from individual developer experimentation to a platform-level.
HUMAIN unveiled humain-m3, a 428B-parameter Arabic model built on MiniMax-M3, at LEAP Riyadh on September 3. The research preview via HUMAIN Node gives developers early access to a model that claims the highest average score across seven Arabic benchmarks.
NVIDIA's BioNeMo Agent Toolkit now runs inside Anthropic's Claude Science, letting AI agents orchestrate protein structure prediction with MSA Search, OpenFold3, and Boltz-2 NIM microservices. The benchmark shows paired MSA input is essential — interface confidence collapses from ~0.85 to ~0.15 without evolutionary alignment.
NVIDIA and CrowdStrike validated a closed-loop agentic cybersecurity system where red and blue agents iterate at machine speed using Nemotron models. The open-model pipeline achieved a 41.9% mean detection rate in backtesting and generalized to 45% of unseen live-fire attacks, outperforming a proprietary frontier system.