Qwen3-Next-80B-A3B-Thinking

Qwen3-Next-80B-A3B-Thinking is a reasoning-focused large language model from Alibaba's Qwen team that always operates in thinking mode, generating a step-by-step reasoning trace before producing its final answer. It uses a hybrid Gated DeltaNet and Gated Attention architecture with 80B total parameters but only 3B active per token, and supports a native 262K-token context window extensible to approximately 1M tokens. Source: supabase.