Understanding Amazon Bedrock mannequin lifecycle

Amazon Bedrock repeatedly releases new basis mannequin (FM) variations with higher capabilities, accuracy, and security. Understanding the mannequin lifecycle is crucial for efficient planning and administration of AI purposes constructed on Amazon Bedrock. Earlier than migrating your purposes, you’ll be able to check these fashions by means of the Amazon Bedrock console or API to judge their efficiency and compatibility.

This submit reveals you the best way to handle FM transitions in Amazon Bedrock, so you can also make certain your AI purposes stay operational as fashions evolve. We focus on the three lifecycle states, the best way to plan migrations with the brand new prolonged entry function, and sensible methods to transition your purposes to newer fashions with out disruption.

Amazon Bedrock mannequin lifecycle overview

A mannequin supplied on Amazon Bedrock can exist in one in all three states: Lively, Legacy, or Finish-of-Life (EOL). Their present standing is seen each on the Amazon Bedrock console and in API responses. For instance, once you make a GetFoundationModel or ListFoundationModels name, the state of the mannequin shall be proven within the modelLifecycle area within the response.

The next diagram illustrates the main points round every mannequin state.

The state particulars are as follows:

• ACTIVE – Lively fashions obtain ongoing upkeep, updates, and bug fixes from their suppliers. Whereas a mannequin is Lively, you need to use it for inference by means of APIs like InvokeModel or Converse, customise it (if supported), and request quota will increase by means of AWS Service Quotas.
• LEGACY – When a mannequin supplier transitions a mannequin to Legacy state, Amazon Bedrock will notify clients with at the very least 6 months’ advance discover earlier than the EOL date, offering important time to plan and execute a migration to newer or different mannequin variations. Through the Legacy interval, current clients can proceed utilizing the mannequin, although new clients is likely to be unable to entry it, and current clients may lose entry for inactive accounts if they don’t name the mannequin for a interval of 15 days or extra. Organizations ought to be aware that creating new provisioned throughput by mannequin models turns into unavailable, and mannequin customization capabilities may face restrictions. For fashions with EOL dates after February 1, 2026, Amazon Bedrock introduces an extra section throughout the Legacy state:
  • Public prolonged entry interval – After spending a minimal of three months in Legacy standing, the mannequin enters this prolonged entry section. Lively customers can proceed utilizing it for at the very least one other 3 months till EOL. Throughout prolonged entry, quota improve requests by means of AWS Service Quotas usually are not anticipated to be authorised, so plan your capability wants earlier than a mannequin enters this section. Throughout this era, pricing could also be adjusted (see Pricing throughout prolonged entry under), and clients will obtain notifications in regards to the transition date and any modifications.
• END-OF-LIFE (EOL) – When a mannequin reaches its EOL date, it turns into fully inaccessible throughout all AWS Areas except particularly famous within the EOL listing. API requests to EOL fashions will fail, rendering them unavailable to most clients except particular preparations exist between the client and supplier for continued entry. The transition to EOL requires proactive buyer motion—migration doesn’t occur mechanically. Organizations should replace their software code to make use of different fashions earlier than the EOL date arrives. When EOL is reached, the mannequin turns into fully inaccessible for many clients.

After a mannequin launches on Amazon Bedrock, it stays obtainable for at the very least 12 months after launch and stays in Legacy state for at the very least 6 months earlier than EOL. This timeline helps clients plan migrations with out speeding.

Pricing throughout prolonged entry

Through the prolonged entry interval, pricing could also be adjusted by the mannequin supplier. If pricing modifications are deliberate, you can be notified within the preliminary legacy announcement and earlier than any subsequent modifications take impact, so there shall be no shock retroactive value will increase. Clients with current non-public pricing agreements with mannequin suppliers or these utilizing provisioned throughput will proceed to function beneath their present pricing phrases throughout the prolonged entry interval. This makes certain clients who’ve made particular preparations with mannequin suppliers or invested in provisioned capability won’t be unexpectedly affected by any pricing modifications.

Communication Course of for Mannequin State Adjustments

Clients will obtain a notification 6 months previous to a mannequin’s EOL date when the mannequin supplier transitions a mannequin to Legacy state. This proactive communication method ensures that clients have ample time to plan and execute their migration methods earlier than a mannequin turns into EOL.

Notifications embody particulars in regards to the mannequin being deprecated, essential dates, prolonged entry availability, and when the mannequin shall be EOL. AWS makes use of a number of channels to make sure these essential communications attain the appropriate folks, together with:

• E-mail notifications
• AWS Well being Dashboard
• Alerts within the Amazon Bedrock console
• Programmatic entry by means of the API.

To be sure to obtain these notifications, confirm and configure your account contact electronic mail addresses. By default, notifications are despatched to your account’s root person electronic mail and alternate contacts (operations, safety, and billing). You possibly can assessment and replace these contacts in your AWS Account web page within the Alternate contacts part. So as to add further recipients or supply channels (reminiscent of Slack or electronic mail distribution lists), go to the AWS Consumer Notifications console and select AWS managed notifications subscriptions to handle your supply channels and account contacts. If you’re not receiving anticipated notifications, test that your electronic mail addresses are accurately configured in these settings and that notification emails from well being@aws.com usually are not being filtered by your electronic mail supplier.

Migration methods and finest practices

When migrating to a more recent mannequin, replace your software code and test that your service quotas can deal with anticipated quantity. Planning forward helps you transition easily with minimal disruption.

Planning your migration timeline

Begin planning as quickly as a mannequin enters Legacy state:

• Evaluation section – Consider your present utilization of the legacy mannequin, together with which purposes rely upon it, typical request patterns, and particular behaviors or outputs that your purposes depend on.
• Analysis section – Examine the really helpful substitute mannequin, understanding its capabilities, variations from the legacy mannequin, new options that might improve your purposes, and the brand new mannequin’s Regional availability. Overview API modifications and documentation.
• Testing section – Conduct thorough testing with the brand new mannequin and examine efficiency metrics between fashions. This helps determine changes wanted in your software code or immediate engineering.
• Migration section – Implement modifications utilizing a phased deployment method. Monitor system efficiency throughout transition and keep rollback functionality.
• Operational section – After migration, repeatedly monitor your purposes and person suggestions to ensure they’re performing as anticipated with the brand new mannequin.

Technical migration steps

Check your migration completely:

• Replace API references – Modify your software code to reference the brand new mannequin ID. For instance, altering from anthropic.claude-3-5-sonnet-20240620-v1:0 to anthropic.claude-sonnet-4-5-20250929-v1:0 or world cross-Area inference world.anthropic.claude-sonnet-4-5-20250929-v1:0. Replace immediate constructions in response to new mannequin’s finest practices. For extra detailed steering, check with Migrate from Anthropic’s Claude Sonnet 3.x to Claude Sonnet 4.x on Amazon Bedrock.
• Request quota will increase – Earlier than totally migrating, be sure to have ample quotas for the brand new mannequin by requesting will increase by means of the AWS Service Quotas console if obligatory.
• Regulate prompts – Newer fashions may reply in a different way to the identical prompts. Overview and refine your prompts accordingly to the brand new mannequin specs. You too can use instruments such because the immediate optimizer in Amazon Bedrock to help with rewriting your immediate for the goal mannequin.
• Replace response dealing with – If the brand new mannequin returns responses in a unique format or with totally different traits, replace your parsing and processing logic accordingly.
• Optimize token utilization – Reap the benefits of effectivity enhancements in newer fashions by reviewing and optimizing your token utilization patterns. For instance, fashions that help immediate caching can cut back the associated fee and latency of your invocations.

Testing methods

Thorough testing is essential for a profitable migration:

• Facet-by-side comparability – Run the identical requests towards each the legacy and new fashions to check outputs and determine any variations that may have an effect on your software. For manufacturing environments, take into account shadow testing—sending duplicate requests to the brand new mannequin alongside your current mannequin with out affecting end-users. With this method, you’ll be able to consider mannequin efficiency, latency and errors charges, and different operational elements earlier than full migration. Carry out A/B testing for person impression evaluation by routing a managed share of stay visitors to the brand new mannequin whereas monitoring key metrics reminiscent of person engagement, activity completion charges, satisfaction scores, and enterprise KPIs.
• Efficiency testing – Measure response occasions, token utilization, and different efficiency metrics to grasp how the brand new mannequin performs in comparison with the legacy model. Validate business-specific success metrics.
• Regression and edge case testing – Ensure that current performance continues to work as anticipated with the brand new mannequin. Pay particular consideration to uncommon or complicated inputs that may reveal variations in how the fashions deal with difficult situations.

Conclusion

The mannequin lifecycle coverage in Amazon Bedrock offers you clear phases for managing FM evolution. Transition intervals provide prolonged entry choices, and provisions for fine-tuned fashions show you how to steadiness innovation with stability.

Keep knowledgeable about mannequin states by means of the AWS Well being Dashboard, plan migrations when fashions enter the Legacy state, and check newer variations completely. These tips may help you keep continuity in your AI purposes whereas utilizing improved capabilities in newer fashions.

In case you have additional questions or issues, attain out to your AWS crew. We wish to show you how to and facilitate a clean transition as you proceed to make the most of the newest developments in FM know-how.

For continued studying and implementation help, discover the official AWS Bedrock documentation for complete guides and API references. Moreover, go to the AWS Machine Studying Weblog and AWS Structure Middle for real-world case research, migration finest practices, and reference architectures that may assist optimize your mannequin lifecycle administration technique.

In regards to the authors

Saurabh Trikande is a Senior Product Supervisor for Amazon Bedrock and Amazon SageMaker Inference. He’s keen about working with clients and companions, motivated by the objective of democratizing AI. He focuses on core challenges associated to deploying complicated AI purposes, inference with multi-tenant fashions, price optimizations, and making the deployment of generative AI fashions extra accessible. In his spare time, Saurabh enjoys mountaineering, studying about progressive applied sciences, following TechCrunch, and spending time along with his household.

Melanie Li, PhD, is a Senior Generative AI Specialist Options Architect at AWS based mostly in Sydney, Australia, the place her focus is on working with clients to construct options utilizing state-of-the-art AI/ML instruments. She has been actively concerned in a number of generative AI initiatives throughout APJ, harnessing the facility of LLMs. Previous to becoming a member of AWS, Dr. Li held knowledge science roles within the monetary and retail industries.

Derrick Choo is a Senior Options Architect at AWS who accelerates enterprise digital transformation by means of cloud adoption, AI/ML, and generative AI options. He focuses on full-stack growth and ML, designing end-to-end options spanning frontend interfaces, IoT purposes, knowledge integrations, and ML fashions, with a selected concentrate on laptop imaginative and prescient and multi-modal programs.

Jared Dean is a Principal AI/ML Options Architect at AWS. Jared works with clients throughout industries to develop machine studying purposes that enhance effectivity. He’s inquisitive about all issues AI, know-how, and BBQ.

Julia Bodia is Principal Product Supervisor for Amazon Bedrock.

Pooja Rao is a Senior Program Supervisor at AWS, main quota and capability administration and supporting enterprise growth for the Bedrock Go-To-Market crew. Outdoors of labor, she enjoys studying, touring, and spending time together with her household.