Arize Model Monitoring is a machine learning operations platform that detects and measures issues in production machine learning models and large language model systems. The platform monitors model performance, data quality, drift detection, and embedding analysis through configurable alerts that trigger when model metrics cross defined thresholds. The platform provides performance analysis monitoring with evaluation metrics including accuracy, recall, precision, F-score, MAE, and MAPE. It enables comparison of model behavior across different environments such as training, validation, and production to identify performance degradation. For drift detection, the platform monitors distribution changes across features, predictions, and actuals. It tracks prediction drift, concept drift, data drift, and upstream drift to identify when model behavior deviates from expected patterns. Data quality monitoring performs real-time checks on features, predictions, and actuals to catch data issues before they impact model performance. The platform includes specialized monitoring capabilities for LLM systems, tracking token usage, latency, and response quality. The platform supports trace instrumentation for LLM applications through OpenInference, enabling collection and visualization of spans. It provides tools for evaluating LLM systems including retrieval augmented generation, prompt engineering, and fine-tuning analysis. Monitoring can be configured across various model types with environment-specific comparisons and cohort-based analysis to drill into root causes of issues.