What are Mixes?
Mixes are a core feature of Crosshatch that address key challenges faced by developers integrating AI capabilities into their applications. They provide access to multiple AI models through a single, unified API, solving several critical problems in AI development:
-
Keeping up with rapidly evolving models: The AI landscape changes quickly, with new and improved models constantly emerging. Mixes ensure developers always have access to top-performing models without needing to continuously update their integrations.
-
Selecting the right model for specific tasks: Different models excel at different tasks, making it challenging to choose the optimal model for each use case. Mixes handle this complexity, automatically selecting or combining models best suited for specific types of requests.
-
Managing multiple API integrations: Traditionally, leveraging multiple AI models required managing several separate API integrations. Mixes simplify this process by providing a single point of integration for accessing various models.
Crosshatch offers two primary types of Mixes:
-
Index Mixes: These automatically route requests to the current top-performing model based on trusted third-party leaderboards. As leaderboard rankings change, the mix updates to ensure access to the best-performing model for the given task.
-
Synthesis Mixes: These combine outputs from multiple models to generate comprehensive responses, aiming to produce higher quality outputs for complex tasks.
Multi-model orchestration, as implemented in Mixes, offers several benefits:
-
Performance Optimization: By leveraging the strengths of multiple models, Mixes can achieve superior performance compared to single-model approaches. For example, the SynthCode Mix has demonstrated an 18% improvement over the current leader in the Bigcodebench Instruct Hard, a benchmark for difficult coding tasks.
-
Robustness: Combining multiple models can lead to more reliable and consistent performance across a wide range of tasks and input types.
-
Adaptability: As new models emerge and performance rankings shift, multi-model approaches can quickly incorporate these changes, ensuring your application always benefits from the latest advancements in AI.
To ensure developers have access to the best AI models, Crosshatch relies on trusted third-party leaderboards, such as those provided by LMSys and Scale.ai. These independent benchmarks offer a more reliable measure of model performance compared to potentially biased or easily gamed internal evaluations.
Mixes represent a significant step forward in simplifying AI integration while maximizing performance. They allow developers to focus on building innovative AI-powered features without getting caught up in the complexities of model selection, integration, and ongoing maintenance.