Claude Platform Updates: New Models, Tooling, and AWS Integration
Anthropic has released updates to its Claude Platform, including new models like Claude Fable 5 and Claude Opus 4.8, offering enhanced capabilities and larger context windows. The platform also introduces new tool functionalities, improved agent orchestration on AWS, and deprecates older models, requiring users to migrate to newer versions. These changes affect developers and organizations using the Claude API and its associated client SDKs and console.
- →Retirement of Claude Sonnet 4 and Opus 4 models
- →Work endpoint for self-hosted sandboxes available on AWS
- →Launch of Claude Fable 5 and Claude Mythos 5 models
- →Code execution tool updated with version 20260521
- →Web search and fetch tools gain response_inclusion parameter
Deprecations (1) ›
- Retirement of Claude Sonnet 4 and Opus 4 models
The Claude Sonnet 4 (claude-sonnet-4-20250514) and Claude Opus 4 (claude-opus-4-20250514) models have been retired. All requests to these models will now result in an error. Users are advised to upgrade to Claude Sonnet 4.6 and Claude Opus 4.8 respectively, with exceptions for researchers through the External Researcher Access Program.
Features (2) ›
- Work endpoint for self-hosted sandboxes available on AWS
The GET /v1/environments/{id}/work endpoint, used for listing pending work in self-hosted sandboxes, is now available on Claude Platform on AWS. Authorization for this action is managed via the GetEnvironment IAM action.
- Launch of Claude Fable 5 and Claude Mythos 5 models
Anthropic has released Claude Fable 5, their most capable widely released model, and Claude Mythos 5 for Project Glasswing participants. Both models feature a 1M token context window by default, 128k max output tokens, and always-on adaptive thinking. Documentation is available for capabilities and API changes.
Enhancements (5) ›
- Code execution tool updated with version 20260521
The code execution tool now supports version code_execution_20260521. This update discloses the 90-second per-cell execution time limit in the tool description, allowing Claude to better budget for long-running cells without requiring a beta header.
- Web search and fetch tools gain response_inclusion parameter
The web search tool (web_search_20260318) and web fetch tool (web_fetch_20260318) have been updated to include a response_inclusion parameter. This allows for the exclusion of consumed result blocks from API responses in agentic workflows, without needing a beta header.
- New tokenizer for Claude Fable 5 and Mythos 5
Claude Fable 5 and Claude Mythos 5 utilize a tokenizer introduced with Claude Opus 4.7. This results in approximately 30% more tokens for the same text compared to previous models. Users can use the token counting API with model: 'claude-fable-5' for accurate prompt measurement.
- Claude Fable 5 safety classifiers and refusal handling
Claude Fable 5 incorporates safety classifiers that return stop_reason: 'refusal' for declined requests, with no billing for refused requests before output generation. An opt-in 'fallbacks' parameter (beta) can re-run refused requests on another model at that model's rates.
- Stop reason field expanded for Claude Fable 5
The 'field' on refusal responses for Claude Fable 5 now includes a category (cyber, bio, or null) and a human-readable explanation. This allows applications to route different classes of refusals appropriately. The existing categories remain unchanged.
https://platform.claude.com/docs/en/release-notes/overview