sparse MoE

Qwen3-Next turned sparse architecture into a long-context test plan

Qwen announced Qwen3-Next on September 10, 2025 with hybrid attention, sparse MoE, and multi-token prediction. Architecture claims need sequence-position and throughput evidence.

Read analysis →