So, you’ve deployed a fantastic application, leveraging the power of SAP‘s Generative AI Hub (linked reference architecture)! That’s a significant step towards building intelligent business processes.

As a quick reminder: SAP’s Generative AI Hub is part of the AI Foundation. SAP’s AI Foundation, similar to an operating system provides the means to build, extend, and run custom AI solutions and agents at scale. It includes the generative AI Hub in SAP AI Core, which is a rich set of AI capabilities (40+ frontier AI models, orchestration, SDKs, etc.), for productive purpose, running on SAP Business Technology Platform (BTP).

Your application runs on BTP, and LLM technology powers the intelligence behind it. But here’s what you can’t delegate: managing the LLM’s Lifecycle for your application.

Understanding Model Lifecycle: A Key Distinction

Before diving into the model lifecycle, let’s clarify the different ways you can leverage AI models:

Your self-hosted Models: You bring your own models (e.g., fine-tuned open-source models, proprietary models) and deploy them. This is akin to a Platform-as-a-Service (PaaS) offering, where you manage the model artifacts, their dependencies, and their whole lifecycle on your own.  This gives you maximum flexibility: you can select, customize, and optimize models to fit your exact needs, including full control over versioning and performance. However, with that freedom comes responsibility. You’re in charge of the full lifecycle — from infrastructure and scaling to security, compliance, monitoring, costs and updates.SAP Managed Models: This is comparable to a Software-as-a-Service (SaaS) offering. The Generative AI Hub provides pre-integrated, managed access to various LLMs from leading providers (i.e Azure OpenAI GPT-5, Google Gemini 2.5 Pro, AWS Claude Sonnet 4, etc.). It’s important to understand that you do not need to acquire separate licenses for these LLMs; SAP handles the commercial agreements and technical integration. Furthermore, SAP takes on the responsibility for data privacy and security or confidentiality standards, related to the LLMs consumed through the Generative AI Hub, significantly reducing your operational burden.

For the following we focus on the latter. LLMs, the powerful engines behind generative AI, are constantly evolving. Model providers release new versions, older versions might be retired, and factors like performance and cost can change, as indicated by the graphic below. This dynamic environment means that the specific LLM version your application relies on today might not be available tomorrow.

Why Does This Matter to You, the Application Owner?

Here’s the reality: AI models get retired. Your application depends on them. Without a lifecycle management strategy, you’re one model retirement away from downtime —and missing out on better quality from newer models.

Moreover, different model versions may offer significantly better performance. A newer model might produce more accurate summaries which can directly impact your application’s quality, user satisfaction, and business outcomes. Not updating could mean missing out on substantial improvements in capability, efficiency, or cost-effectiveness.

Your Role: Stay Informed, Be Prepared

Your primary responsibility regarding the model lifecycle is to:

Stay informed about the models supported by the SAP’s Generative AI Hub, including their versions, capabilities, cost, and most importantly, their retirement dates.Have a plan for upgrading your application’s model configuration when necessary.

Let’s have a look on how you can fulfill these responsibilities using the tools and information available.

Staying Informed: Where to Find Model Information

SAP’s Generative AI Hub provides several ways to discover and understand the available LLMs:

SAP Notes:

The central source of truth for supported models is SAP Note 3437766. This note provides comprehensive information, including the list of available models, supported versions, token conversion rates, rate limits, and retirement dates. Regularly checking this note is essential. You can access it via https://me.sap.com/notes/3437766. Crucially, you can subscribe to this SAP Note (indicated by the star icon) to receive automatic notifications via email about any changes, ensuring you’re always up-to-date.

SAP AI Launchpad (UI):

It’s the natural place for an application owner to explore the Model Library and discover available models and their details. The Model Library also supports through comparison functionalities like the Leaderboard tab. Here you can see similar or better performing models.

Discovery API Endpoint:

For programmatic access or detailed inspection, you can use the discovery API endpoint. By sending a GET request to the {{apiurl}}/v2/lm/scenarios/{scenarioid}/models endpoint, you can retrieve a list of all available models, their versions, capabilities, and retirement dates. This is a powerful tool for checking information dynamically. Find further information about that on the SAP Business Accelerator Hub.

By utilizing these resources, you can proactively identify when the model your application relies on is approaching its retirement date and plan your next steps.

Taking Action: Upgrading Your Model

When it’s time to switch to a newer model version (preceded by tests/benchmarks), you have two main strategies, decided during the initial deployment configuration:

Auto Upgrade (modelVersion: latest):

How it works: When you create your generative AI configuration for a specific LLM, either through orchestration API or as a deployment, you can set the modelVersion parameter to latest. This will automatically use the most recent version of the specified model that is supported.Pros: Less manual effort. As we support newer versions, your deployment will automatically shift, potentially giving you access to model improvements without intervention.Cons: Less control. Your application might suddenly start using a new model version which, although intended to be compatible, could introduce subtle changes in behavior or performance that you haven’t explicitly tested. This can be particularly problematic if your application relies on consistent LLM output behavior. Furthermore, it won’t help in the situation of moving to different models (i.e. GPT-4.1 to GPT-5).

Manual Upgrade (Specify <modelVersion>):

How it works: When creating your orchestration configuration, you specify a particular modelVersion, like “2024-05-13”. When this specific version is close to retirement, you will need to:Select a suitable newer model version based on your benchmarks using the SAP Note, Discovery API and/or UI.Via our harmonized API in orchestration, it is easy for you to change your configuration to a newer, specific model version.

And only if you handle your custom defined endpoint:

Patch upgrade your existing deployment to the configurationId you created newly.Pros: Full control. You decide exactly which model version your application uses and when to make the change, allowing for thorough testing beforehand. This is especially important for production systems where behavioral changes in LLMs can have significant impacts.Cons: More manual effort. You must actively monitor retirement dates and perform the upgrade steps yourself.

Choosing Your Strategy: Best Practices

The best strategy depends on your application’s needs for stability versus automatically receiving updates.

For most production applications, especially those sensitive to potential behavioral changes in LLMs, specifying a fixed modelVersion (Manual Upgrade) has emerged as a best practice. While it requires more manual effort, it provides the necessary control and predictability. Most application owners prefer not to use the latest parameter due to the risk of unexpected behavior shifts with new LLM versions. By fixing the version, you safeguard against such changes and can plan updates proactively.

If stability and controlled changes are paramount, opt for Manual Upgrade by specifying a fixed modelVersion and plan for scheduled updates. This allows you to thoroughly test new model versions in a controlled environment before deploying them to production.

If minimizing operational overhead and automatically getting the latest model version is preferred for non-critical or exploratory applications, use Auto Upgrade with modelVersion: latest. However, remain aware that behavior might change with new versions and you need to change from model to model.

In Summary

As an application owner, understanding and managing the LLM model lifecycle is key to your application’s longevity and reliability.

Regularly consult SAP Note 3437766 and/or the Discovery API (or the UI) to track model versions and retirement dates. Remember to subscribe to the SAP Note for automated updates.Choose an appropriate upgrade strategy (latest for automatic updates or specifying a modelVersion for manual control) when setting up your configuration and deployment. For production scenarios, specifying a fixed modelVersion is generally recommended as a best practice.If using manual upgrades, proactively change your configuration within orchestration and if custom deployment is used you need to patch/upgrade the current model version before it expires.

By staying informed and prepared, you can ensure your generative AI application continues to deliver value seamlessly, leveraging the latest innovations from the SAP’s Generative AI Hub.

We recommend keeping a close eye on the What’s new section for upcoming release announcements and implementation guidance.

​ So, you’ve deployed a fantastic application, leveraging the power of SAP‘s Generative AI Hub (linked reference architecture)! That’s a significant step towards building intelligent business processes.As a quick reminder: SAP’s Generative AI Hub is part of the AI Foundation. SAP’s AI Foundation, similar to an operating system provides the means to build, extend, and run custom AI solutions and agents at scale. It includes the generative AI Hub in SAP AI Core, which is a rich set of AI capabilities (40+ frontier AI models, orchestration, SDKs, etc.), for productive purpose, running on SAP Business Technology Platform (BTP).Your application runs on BTP, and LLM technology powers the intelligence behind it. But here’s what you can’t delegate: managing the LLM’s Lifecycle for your application. Understanding Model Lifecycle: A Key DistinctionBefore diving into the model lifecycle, let’s clarify the different ways you can leverage AI models:Your self-hosted Models: You bring your own models (e.g., fine-tuned open-source models, proprietary models) and deploy them. This is akin to a Platform-as-a-Service (PaaS) offering, where you manage the model artifacts, their dependencies, and their whole lifecycle on your own.  This gives you maximum flexibility: you can select, customize, and optimize models to fit your exact needs, including full control over versioning and performance. However, with that freedom comes responsibility. You’re in charge of the full lifecycle — from infrastructure and scaling to security, compliance, monitoring, costs and updates.SAP Managed Models: This is comparable to a Software-as-a-Service (SaaS) offering. The Generative AI Hub provides pre-integrated, managed access to various LLMs from leading providers (i.e Azure OpenAI GPT-5, Google Gemini 2.5 Pro, AWS Claude Sonnet 4, etc.). It’s important to understand that you do not need to acquire separate licenses for these LLMs; SAP handles the commercial agreements and technical integration. Furthermore, SAP takes on the responsibility for data privacy and security or confidentiality standards, related to the LLMs consumed through the Generative AI Hub, significantly reducing your operational burden.For the following we focus on the latter. LLMs, the powerful engines behind generative AI, are constantly evolving. Model providers release new versions, older versions might be retired, and factors like performance and cost can change, as indicated by the graphic below. This dynamic environment means that the specific LLM version your application relies on today might not be available tomorrow. Why Does This Matter to You, the Application Owner?Here’s the reality: AI models get retired. Your application depends on them. Without a lifecycle management strategy, you’re one model retirement away from downtime —and missing out on better quality from newer models.Moreover, different model versions may offer significantly better performance. A newer model might produce more accurate summaries which can directly impact your application’s quality, user satisfaction, and business outcomes. Not updating could mean missing out on substantial improvements in capability, efficiency, or cost-effectiveness.Your Role: Stay Informed, Be PreparedYour primary responsibility regarding the model lifecycle is to:Stay informed about the models supported by the SAP’s Generative AI Hub, including their versions, capabilities, cost, and most importantly, their retirement dates.Have a plan for upgrading your application’s model configuration when necessary.Let’s have a look on how you can fulfill these responsibilities using the tools and information available.Staying Informed: Where to Find Model InformationSAP’s Generative AI Hub provides several ways to discover and understand the available LLMs:SAP Notes:The central source of truth for supported models is SAP Note 3437766. This note provides comprehensive information, including the list of available models, supported versions, token conversion rates, rate limits, and retirement dates. Regularly checking this note is essential. You can access it via https://me.sap.com/notes/3437766. Crucially, you can subscribe to this SAP Note (indicated by the star icon) to receive automatic notifications via email about any changes, ensuring you’re always up-to-date.SAP AI Launchpad (UI):It’s the natural place for an application owner to explore the Model Library and discover available models and their details. The Model Library also supports through comparison functionalities like the Leaderboard tab. Here you can see similar or better performing models.Discovery API Endpoint:For programmatic access or detailed inspection, you can use the discovery API endpoint. By sending a GET request to the {{apiurl}}/v2/lm/scenarios/{scenarioid}/models endpoint, you can retrieve a list of all available models, their versions, capabilities, and retirement dates. This is a powerful tool for checking information dynamically. Find further information about that on the SAP Business Accelerator Hub.By utilizing these resources, you can proactively identify when the model your application relies on is approaching its retirement date and plan your next steps.Taking Action: Upgrading Your ModelWhen it’s time to switch to a newer model version (preceded by tests/benchmarks), you have two main strategies, decided during the initial deployment configuration:Auto Upgrade (modelVersion: latest):How it works: When you create your generative AI configuration for a specific LLM, either through orchestration API or as a deployment, you can set the modelVersion parameter to latest. This will automatically use the most recent version of the specified model that is supported.Pros: Less manual effort. As we support newer versions, your deployment will automatically shift, potentially giving you access to model improvements without intervention.Cons: Less control. Your application might suddenly start using a new model version which, although intended to be compatible, could introduce subtle changes in behavior or performance that you haven’t explicitly tested. This can be particularly problematic if your application relies on consistent LLM output behavior. Furthermore, it won’t help in the situation of moving to different models (i.e. GPT-4.1 to GPT-5).Manual Upgrade (Specify <modelVersion>):How it works: When creating your orchestration configuration, you specify a particular modelVersion, like “2024-05-13”. When this specific version is close to retirement, you will need to:Select a suitable newer model version based on your benchmarks using the SAP Note, Discovery API and/or UI.Via our harmonized API in orchestration, it is easy for you to change your configuration to a newer, specific model version.And only if you handle your custom defined endpoint:Patch upgrade your existing deployment to the configurationId you created newly.Pros: Full control. You decide exactly which model version your application uses and when to make the change, allowing for thorough testing beforehand. This is especially important for production systems where behavioral changes in LLMs can have significant impacts.Cons: More manual effort. You must actively monitor retirement dates and perform the upgrade steps yourself.Choosing Your Strategy: Best PracticesThe best strategy depends on your application’s needs for stability versus automatically receiving updates.For most production applications, especially those sensitive to potential behavioral changes in LLMs, specifying a fixed modelVersion (Manual Upgrade) has emerged as a best practice. While it requires more manual effort, it provides the necessary control and predictability. Most application owners prefer not to use the latest parameter due to the risk of unexpected behavior shifts with new LLM versions. By fixing the version, you safeguard against such changes and can plan updates proactively.If stability and controlled changes are paramount, opt for Manual Upgrade by specifying a fixed modelVersion and plan for scheduled updates. This allows you to thoroughly test new model versions in a controlled environment before deploying them to production.If minimizing operational overhead and automatically getting the latest model version is preferred for non-critical or exploratory applications, use Auto Upgrade with modelVersion: latest. However, remain aware that behavior might change with new versions and you need to change from model to model. In SummaryAs an application owner, understanding and managing the LLM model lifecycle is key to your application’s longevity and reliability.Regularly consult SAP Note 3437766 and/or the Discovery API (or the UI) to track model versions and retirement dates. Remember to subscribe to the SAP Note for automated updates.Choose an appropriate upgrade strategy (latest for automatic updates or specifying a modelVersion for manual control) when setting up your configuration and deployment. For production scenarios, specifying a fixed modelVersion is generally recommended as a best practice.If using manual upgrades, proactively change your configuration within orchestration and if custom deployment is used you need to patch/upgrade the current model version before it expires.By staying informed and prepared, you can ensure your generative AI application continues to deliver value seamlessly, leveraging the latest innovations from the SAP’s Generative AI Hub.We recommend keeping a close eye on the What’s new section for upcoming release announcements and implementation guidance.     Read More Technology Blog Posts by SAP articles 

#SAP

#SAPTechnologyblog