Qwen3.7 Max

Qwen3.7 Max shows slight performance decline across key benchmarks

Qwen3.7 Max has experienced measurable performance decreases across multiple evaluation categories in the latest benchmark window. The model's overall score dropped from 68.1 to 66.3, reflecting a broader pattern of declining capabilities. Coding performance, previously a standout strength, fell from 72.9 to 70.5, while instruction following decreased from 77.7 to 75.8. Mathematical reasoning also declined from 71.9 to 70.3, and creative writing saw a reduction from 53.8 to 51.9. The model maintained its position in hard prompts with a modest improvement from 59.8 to 60.2, suggesting some stability in handling complex tasks. Style control remained relatively steady at 74.4 compared to the previous 75.0. Speed metrics show the model continues to operate quickly with a time to first token of 0.58 seconds and total time of 10.79 seconds. Despite these performance regressions, Qwen3.7 Max remains a capable model particularly for coding applications, though users should be aware of the downward trajectory in several key competency areas. The changes suggest potential model updates or infrastructure adjustments that have affected output quality.