v3.11.0rc1

The MLflow project, a popular open-source platform for managing the machine learning lifecycle, has pushed a significant pre-release update with version 3.11.0rc1. The core of this release is a major architectural shift for its AI Gateway and evaluation components. The development team has systematically stripped out third-party dependencies that were previously used to route requests to external AI providers. In their place, MLflow now implements its own native routing logic, creating a more self-contained and controllable system for interacting with models from providers like OpenAI, Anthropic, and others.

This change has substantial implications for developers and ML engineers. By eliminating external libraries, MLflow reduces its attack surface, mitigating potential security risks that can come from transitive dependencies. It also simplifies the deployment and dependency management process, as there are fewer external packages to version-lock and manage. For teams using the AI Gateway to standardize access to multiple LLMs, this built-in approach offers greater transparency and control over the request/response lifecycle, potentially improving reliability and making debugging easier. The move signals MLflow's commitment to maturing its AI-centric features into a robust, first-class component of its MLOps suite.