Qwen: Qwen3 Next 80B A3B Thinking

Qwen3-Next-80B-A3B-Thinking, created on September 11, 2025, is a reasoning-first chat model designed for complex multi-step problems. It operates in a thinking-only mode, outputting structured "thinking" traces by default, making it highly suitable for agent frameworks, tool use, and retrieval-heavy workflows. The model exhibits exceptional reliability, achieving a 100% success rate across all benchmarks, indicating consistent and usable responses. However, it tends to have longer response times, ranking in the 8th percentile for speed, and is positioned at premium pricing levels, ranking in the 5th percentile for cost. In terms of performance, Qwen3-Next-80B-A3B-Thinking demonstrates significant strengths in specialized areas. It achieves outstanding accuracy in Coding (94.9%), Reasoning (96.0%), and Ethics (100%), with the Ethics benchmark highlighting it as the most accurate model at its price point and among models of similar speed. Its General Knowledge (99.5%) and Email Classification (99.0%) capabilities are also strong. A notable weakness is its Instruction Following, where it scored only 14.7% accuracy, indicating challenges with highly complex, multi-layered instructions despite its design for structured thinking. This suggests that while it excels at generating detailed thought processes, translating those into precise, multi-faceted instruction adherence remains an area for improvement.