As enterprises increasingly adopt AI-driven solutions, managing and selecting the right machine learning and generative AI models becomes critical. The Model Library in SAP AI Launchpad provides a centralized interface to streamline this process, and one of its most powerful features.
The Model Library acts as a catalog of available Generative AI models, allowing developers, data scientists, and architects to explore, evaluate, and integrate models into enterprise workflows efficiently.
What is the Model Library?
The Model Library in SAP AI Launchpad is a centralized repository of Generative AI models from multiple providers (such as Anthropic, OpenAI, SAP, etc.), enriched with:
- Model metadata
- Capabilities and modalities
- Versioning details
- Benchmarking insights (via leaderboard and charts)
- Integration readiness
It essentially serves as a single pane of glass for model discovery and selection. Following SAP note 3437766 – Availability of Generative AI Models outlines the available models and their versions.
Key Components of the Model Library UI
The Model Library in SAP AI Launchpad is structured into several key sections:
Filters
The Model Library offers several filter categories to help narrow down their model selection.
- Capabilities lets users filter by what a model can do, such as embedding, text generation, speech to text, image understanding, or reasoning.
- Input Types allows filtering based on the kinds of inputs a model accepts, like audio, text, video, images, or documents (pdf, txt).
- Model Provisioning helps users find models based on how they are deployed or made available within the platform such as SAP Hosted, SAP Managed and Remote.
- Model Provider enables filtering by the organization behind the model, such as Anthropic, Cohere, or OpenAI.
- Access Type allows to find which model are accessed through directly model deployment or through orchestration deployment.
- Other Filters provide additional controls like Latest Version, Streaming Support for refining model selection.
Together, these filters make it easier to identify the right model for a specific use case out of the 45 available in the library as of writing this blog post
Catalog Mode
The Catalog mode presents each model as an individual “Model Card,” providing a clean, tile-based overview of your available AI inventory. This layout is designed for quick scanning and selection.
Each card in this catalog highlights several critical details:
- Model Name and Provider: At the top of each card, you can clearly see the model’s identity (e.g., Claude 4.5 Opus, GPT-4.1 mini, or Cohere Rerank) alongside the provider logo.
- Technical Identifier: Beneath the name is the specific model string (like
anthropic--claude-4.5-sonnet) used for API calls. - Version & Release Date: It tracks the specific version number and, in many cases, the exact release date, which is vital for maintaining consistency in your applications.
- Capability Icons: The small gray icons at the bottom of each card represent the model’s “skills”—such as Text Generation (the “T” icon), Reasoning, or Vision—allowing you to see at a glance if a model supports your specific input type.
Leaderboard Mode
The Leaderboard view in Model Library enables comparison of models across:
- Benchmark scores
- Accuracy
- Latency
- Cost efficiency
The following are the different performance/safety benchmarks on which foundational models are compared
- HELM Lite Accuracy Mean Win Rate
- ChatBotArena Arena Score
- AIRBench Refusal Rate
- AIRBench Discrimination/Bias Refusal Rate
- LMArena Text Arena Score
- HELM Capabilities Accuracy Mean Score
Along with benchmark metrics it also provides additional technical and cost metrics like
- Output Token Cost (Capacity Units)
- Input Token Cost (Capacity Units)
- Context Window
Chart Mode
The Chart mode in the Model Library provides a multi-dimensional visualization designed to help you compare Foundation Models across different performance benchmarks and technical specifications simultaneously.
For example by plotting models on a scatter plot, it allows you to analyze trade-offs between two key variables—such as the ChatBotArena Arena Score on the x-axis and the Context Window size on the y-axis—making it easier to identify “frontier” models that offer the best balance of reasoning capability and data capacity.
Each data point is color-coded by the specific model version (e.g., Claude, GPT-4.1, Gemini), enabling you to quickly spot clusters of high-performing models or identify outliers that might offer exceptionally large context windows at lower performance tiers.
