Tag: mixture-of-experts

All the articles with the tag "mixture-of-experts".

• AI Signals

  Kimi K3 Is Here: What the 2.8T Open Model Actually Changes

  22 Jul, 2026

  Moonshot's Kimi K3 combines 2.8T parameters, sparse MoE routing, native vision, and a 1M context window. Agent reliability and serving cost remain open.

• AI Engineering

  Google Made Qwen 397B Up to 4.7x Faster Without Training a New Model

  15 Jul, 2026

  Google's Ironwood optimization work shows how sharding, fused kernels, and memory-aware serving can change the economics of a frontier-scale open-weight model.