Anthropic: Scheduled deployments for agents and Swift package — key developer news
Alongside the launch of Fable 5 and Mythos 5, Anthropic announced several significant developer platform updates: Claude Managed Agents gain scheduled deployments on a cron schedule, vault support for environment-variable credentials, and a new Swift package that exposes Claude as a server-side LanguageModel in Apple's Foundation Models framework for iOS 27 and macOS 27.
This article was generated using artificial intelligence from primary sources.
Alongside the launch of Claude Fable 5 and Mythos 5, Anthropic announced a series of technical updates for development teams on the Claude platform on June 9, 2026. Two stand out as standalone upgrades: scheduled deployments for Claude Managed Agents and a new Swift package for the Apple ecosystem.
Scheduled deployments: agents without a custom scheduler
Claude Managed Agents now support scheduled deployments — the ability to trigger agentic sessions on a cron schedule without managing your own scheduler infrastructure.
Until now, developers who wanted to run agents at regular intervals had to build their own orchestration logic — Lambda functions, cron jobs, or similar infrastructure. Scheduled deployments move that responsibility to the platform level: the developer defines a cron expression, and the Claude platform autonomously starts sessions at the scheduled time.
This is especially useful for:
- Periodic data processing and reporting
- Automated change monitoring (monitoring agents)
- Nightly batch processing without manual intervention
Alongside scheduled deployments, vaults in Claude Managed Agents now support environment-variable credentials — secure injection of secret keys and credentials into the agent’s sandbox environment as environment variables. This allows CLI tools, SDKs, and services that use environment variables for authentication to integrate directly without code changes.
What is new with the API for Fable 5?
New model IDs and tokenizer
The models claude-fable-5 and claude-mythos-5 are now available (Mythos 5 for Project Glasswing partners only). Both use the tokenizer introduced with Claude Opus 4.7 — the same text generates approximately 30% more tokens than on models predating Opus 4.7. To measure prompts under the new tokenizer, the token-counting API accepts model: "claude-fable-5".
Refusal mechanics and fallbacks
When a Fable 5 classifier refuses a request, the Messages API returns stop_reason: "refusal". A request refused before generating any output is not billed.
The optional fallbacks parameter (in beta on the Claude API and Claude Platform on AWS; not supported for the Message Batches API) automatically retries the refused request on an alternative model, billed at that model’s rates.
New refusal category: reasoning_extraction
The stop_details.category field now includes a third value: "reasoning_extraction" — blocks requests that could be used to distill or replicate Fable 5 capabilities in violation of Anthropic’s Terms of Service. The existing categories "cyber" and "bio" remain unchanged. No beta header is required for this functionality.
Thinking mode restrictions
On Fable 5 and Mythos 5, adaptive thinking is the only supported thinking mode:
thinking: {"type": "disabled"}returns a 400 error — it cannot be disabled- Manual extended thinking budgets and assistant prefill are not supported (return 400)
thinking.displaydefaults to"omitted"— same as Opus 4.8, Opus 4.7, and Mythos Preview- Setting
display: "summarized"returns readable thinking summaries - Raw chain-of-thought is never returned; thinking blocks should be forwarded unchanged in multi-turn conversations
Data retention
Fable 5 requires 30-day data retention on the Claude API and is not available under zero data retention.
New: session_thread_id in webhook events
session.thread_* webhook events now include a session_thread_id field that identifies the multi-agent thread that triggered the event. This makes it easier to track and debug complex multi-agent topologies where multiple agents collaborate within a single context tree.
Swift package: Claude in Apple’s Foundation Models framework
Anthropic has released a Swift package in beta that exposes Claude as a server-side LanguageModel within Apple’s Foundation Models framework.
The package lets you call Claude via the same LanguageModelSession API that iOS developers use for Apple’s on-device model. Key note: Claude operates as a server-side model here — requests go to Anthropic’s infrastructure, not locally on the device — but the developer-facing interface remains identical.
Supported platforms (all in beta):
- iOS 27
- macOS 27
- visionOS 27
- watchOS 27
This is especially interesting for Swift developers already building apps with the Foundation Models API for the on-device model who want to use Claude’s capabilities for more demanding tasks — without changing the interface in their code.
Summary of June 9 developer updates
| Update | Category |
|---|---|
claude-fable-5, claude-mythos-5 available | Models |
| Tokenizer: +30% tokens vs. pre-4.7 models | Tokenization |
stop_reason: "refusal" without billing | API mechanics |
fallbacks parameter (beta) | API mechanics |
stop_details.category: "reasoning_extraction" | Safety |
thinking: disabled returns 400 | API mechanics |
| Scheduled deployments for Managed Agents | Agents |
| Environment-variable credentials in vaults | Agents |
session_thread_id in webhook events | Webhooks |
| Swift package (beta) — Foundation Models | SDK |
All changes listed above take effect immediately for new requests on the Claude API.
Frequently Asked Questions
- What are scheduled deployments in Claude Managed Agents?
- A new capability that allows agentic sessions to be triggered on a cron schedule, without the need to manage a custom scheduler. The developer defines a cron expression and the agent automatically starts sessions at the scheduled time.
- What is the new Claude Swift package and which platforms does it support?
- A beta Swift package that exposes Claude as a server-side LanguageModel in Apple's Foundation Models framework. It is called via the same LanguageModelSession API as Apple's on-device model, available on iOS 27, macOS 27, visionOS 27, and watchOS 27 (beta).
- What happens when Fable 5 refuses a request, and how does the fallbacks parameter work?
- The API returns stop_reason: 'refusal' with no charge if no output was generated. The optional fallbacks parameter (in beta) automatically retries the refused request on another model, billed at that fallback model's rates.
📬 AI news in your inbox
A daily digest built your way — pick topics, sources and cadence. One-click unsubscribe.