Tag: MoE
Qwen3.8-Max’s Open Weights Are Not the Model You Benchmarked
Alibaba opened its 2.4T flagship — but the download is text-only, thinking-required and 262K context, under a licence that is not Apache 2.0.
Nemotron 3.5 Lightning: One GPU, Yes. A Laptop, No.
NVIDIA's 30B Nemotron 3.5 Lightning runs on a single desktop GPU, but its Q4_K_M GGUF is 25.3 GB — above every 24 GB mobile card.
Qwen3.8-Flash-Next Runs Locally in 75 GB of RAM
Alibaba's 125B multimodal MoE is out as Unsloth GGUFs. The 1-bit build fits in about 75 GB of RAM or unified memory, with no GPU VRAM required.
Best Open-Weight LLM for Agentic Coding in 2026
A decision guide to the top open-weight coding models in 2026 — DeepSeek V4, GLM-5.2, Kimi K3, Qwen, and MiniMax M3 — by price, license, and context.
Mixture of Experts Explained: Why Every 2026 Model Uses It
Mixture of Experts lets a model have trillions of parameters but only use a fraction per token. Here's how it works and why it took over in 2026.
Thinking Machines Releases Inkling, Its First Open-Weights Model
Thinking Machines Lab ships Inkling, a 975B-parameter multimodal MoE under Apache 2.0 — the leading U.S. open-weights model at release, per Artificial Analysis.
MiniMax M3 Open-Sourced on Hugging Face
MiniMax has open-sourced its M3 model, a 428B MoE architecture optimized for long context and agentic scenarios, now available on Hugging Face.
Ant Group’s Ling-2.6-flash: Lean MoE for AI Agents
Ant Group releases Ling-2.6-flash, an efficient 104B MoE model optimized for 'intelligence per token,' agentic workflows, and fast long-context performance.
Holo3: H Company’s SOTA Foundation Model for Desktop Agents
H Company unveils Holo3, a high-performance Mixture-of-Experts model family that sets a new industry standard for autonomous desktop application control.
Xiaomi MiMo-V2: Three New State-of-the-Art AI Models
Xiaomi releases MiMo-V2, a groundbreaking AI trio featuring a 1T-parameter Pro model, an omnimodal agent, and a next-gen TTS system.