Little Qwen 3.5 27B and Qwen 35B-A3B models did very well in my logical reasoning benchmark

Alibaba's Qwen AI team is making waves with the performance of its smaller-scale language models. The Qwen 3.5 27B and Qwen 35B-A3B models have delivered impressive results on the lineage-bench, a logical reasoning benchmark that tests a model's ability to draw conclusions from extensive chains of premises. This performance from sub-40B parameter models is notable because logical reasoning has traditionally been a strength reserved for much larger models, often with 70B parameters or more. The achievement suggests significant architectural efficiency and could signal a shift where high-level reasoning becomes more computationally affordable.

The lineage-bench specifically evaluates a model's capacity for reliable, multi-step deduction from hundreds of given facts, a task that challenges both comprehension and logical consistency. For developers and enterprises, the strong showing of these compact Qwen models means the potential for deploying capable reasoning agents at a lower computational cost and with faster inference times. This aligns with the industry trend toward creating more efficient, specialized models rather than simply scaling parameters. The results will likely increase competitive pressure in the mid-size model segment and encourage further benchmarking of reasoning capabilities beyond simple knowledge recall.