QWEN3-NEXT

Summary from the TLDR AI newsletter

Qwen3-Next is a new model architecture featuring a sparse Mixture-of-Experts design with hybrid attention and multi-token prediction. Its 80B-parameter base model activates only 3B parameters at inference, enabling 10x faster throughput for long-context tasks.