Amazon Bedrock introduces extended access phase for legacy models with 6-month deprecation notice
Amazon Bedrock has formalized its model lifecycle management with three states—Active, Legacy, and End-of-Life—guaranteeing at least 12 months of availability after launch and 6 months in Legacy status before deprecation. For models with EOL dates after February 1, 2026, a new extended access phase provides an additional 3+ months of access during Legacy state, giving customers more time to migrate applications.
Amazon Bedrock Introduces Extended Access Phase for Deprecating Models
Amazon has published detailed guidance on its Amazon Bedrock foundation model lifecycle, introducing a new extended access phase that gives customers additional time to migrate applications away from deprecated models.
Model Lifecycle States
Amazon Bedrock models now operate in one of three states, all visible via the console and API responses:
Active – Models receive ongoing maintenance, updates, and bug fixes. Customers can use them for inference, customize them (if supported), and request quota increases through AWS Service Quotas.
Legacy – When a model provider deprecates a model, Amazon Bedrock provides at least 6 months' advance notice before the End-of-Life (EOL) date. Existing customers can continue using Legacy models, though new customers may lose access. Accounts inactive for 15+ days may lose access, and new provisioned throughput creation is unavailable. For models with EOL dates after February 1, 2026, Legacy status now includes a public extended access period.
End-of-Life (EOL) – Models become completely inaccessible across all AWS Regions. API requests fail, and migration is mandatory.
Extended Access Phase
The extended access period is a new sub-phase within Legacy status. After a minimum of 3 months in Legacy state, models enter extended access and remain available for at least another 3 months until EOL. During this phase:
- AWS Service Quotas increase requests are not expected to be approved
- Pricing may be adjusted with advance notification
- Existing provisioned throughput and private pricing agreements continue under current terms
- Active users retain continued access
Minimum Timeline Guarantees
All models remain available for at least 12 months after launch. Models transition through Legacy status for a minimum of 6 months before EOL. For new EOL dates (after February 1, 2026), the extended access phase guarantees an additional 3+ months of access.
Communication and Notifications
Amazon notifies customers 6 months prior to EOL through multiple channels:
- Email notifications (sent to root user and alternate contacts)
- AWS Health Dashboard
- Amazon Bedrock console alerts
- Programmatic API access
Notifications include the deprecated model name, important dates, extended access availability, and EOL date. Customers should verify contact email addresses in their AWS Account settings and configure additional delivery channels (Slack, distribution lists) via the AWS User Notifications console.
Migration Strategy
Amazon recommends a four-phase migration approach:
- Assessment – Evaluate current usage, dependent applications, request patterns, and model-specific behaviors
- Research – Investigate replacement models, capability differences, new features, regional availability, and API changes
- Testing – Conduct thorough testing with the new model, compare performance metrics, and identify required adjustments
- Migration – Implement phased deployment, monitor performance, maintain rollback capability, and continuously monitor post-migration
Technical steps include updating API references (e.g., from anthropic.claude-3-5-sonnet-20240620-v1:0 to anthropic.claude-sonnet-4-5-20250929-v1:0), adjusting prompts for new model behavior, and requesting quota increases before full migration.
What this means
Amazon is standardizing deprecation practices across Bedrock with enforceable timelines and extended access phases. The 6-month minimum Legacy period plus an additional 3+ months of extended access (for new EOLs) gives enterprises roughly 9+ months to migrate—substantially longer than typical cloud service deprecation windows. The multi-channel notification system and emphasis on advance planning reduces surprise EOLs, but organizations still bear responsibility for proactive migration before API failures occur. This is less about new model availability and more about risk management for production workloads.
Related Articles
Meta adds "spaces" to Muse AI, letting users build app-like hubs by describing them
Meta is rolling out "spaces" to all users of its Muse AI assistant. Spaces are customized, app-like experiences that Muse builds from a text description, and they can update automatically when connected to a relevant app. Meta is also releasing a "specialized" version of its Muse Wonder AI model.
Google Antigravity 2.29 adds cloud Remote Agents for Google AI Pro and Ultra subscribers
Google's Antigravity 2.29 adds Remote Agents that run in isolated cloud virtual machines for Google AI Pro and Ultra subscribers. The update also brings a Gemini Live-powered Voice Mode, a redesigned conversation page, and changes to Automations and the built-in browser.
Perplexity launches 'teammates' AI agents inside Computer, exclusive to $200-a-month Max plan
Perplexity has launched 'teammates,' personal AI agents built on Perplexity Computer that handle ongoing work autonomously with minimal supervision. The feature is exclusive to Perplexity Max subscribers, who pay $200 a month or $2,000 a year, on web and iOS.
Jimeng Short Drama Studio links script, assets and storyboards; Seedance 2.5 as low as 0.15 yuan per second
Jimeng (即梦) Short Drama Studio puts scriptwriting, an asset library and storyboarding in a single project for episodic AI video. Qbitai (量子位), in a Chinese-language hands-on test, reports Seedance 2.5 priced as low as 0.15 yuan per second for teams during the Double Eleven promotion.
Comments
Loading...