Qwen: Qwen3.6 27B

Qwen3.6 27B, a 27-billion-parameter multimodal model from Alibaba's Qwen Team, was released in April 2026 with a substantial context length of 262,144. This model exhibits exceptional reliability, boasting a 98% success rate across benchmarks, indicating consistent delivery of usable responses. However, its performance in terms of speed and cost positions it at the lower end of the spectrum; it tends to have longer response times (6th percentile) and is positioned at premium pricing levels (12th percentile). In specific benchmark evaluations, Qwen3.6 27B demonstrates a notable strength in Email Classification, achieving 98.0% accuracy, placing it in the 53rd percentile for this task. Its Instruction Following capabilities are also solid, with 72.7% accuracy (76th percentile). A key area for improvement is its handling of uncertainty, as evidenced by an 80.0% accuracy in the Hallucinations (Baseline) test, ranking it in the 23rd percentile. This suggests the model occasionally struggles to appropriately acknowledge when it lacks knowledge. Overall, Qwen3.6 27B is a highly reliable model with strong instruction following and classification abilities, though its speed, cost, and hallucination mitigation are areas for further development.