“Newer reasoning models, such as recent versions from DeepSeek and Qwen, show a trend of declining performance on simple Question-Answering benchmarks when evaluated without tools.”