To analyze whether modern open-source LLMs are well-calibrated, and how calibration varies across different task types such as reasoning, common sense, binary decision making, and factual truthfulness.