Claude Fable 5 Exits Free Plans Today — API Breaking Changes Developers Must Handle
Summary: Claude Fable 5, launched June 9, ends its complimentary window in Pro, Max, Team, and seat-based Enterprise plans today—billed at API rates going forward—while carrying two API-level changes that require code updates for any existing integration.
Key Facts
- Billing change: Usage now draws from credits at $10/M input and $50/M output tokens — less than half the price of Claude Mythos Preview; refused requests that produce no output are not charged
- New response type: Declined requests return HTTP 200 with
stop_reason: "refusal", not an error; Anthropic offers a server-sidefallbacksparameter (beta) and SDK middleware for automatic retry with fallback credit covering prompt-cache costs - Thinking mode: Adaptive Thinking is always on —
thinking: {"type": "disabled"}is not supported; raw chain-of-thought is never exposed, only a summary or omitted block - Claude Mythos 5 (
claude-mythos-5): Same model, safety classifiers removed, limited to Project Glasswing approved customers; successor to Claude Mythos Preview - Context: 1M-token default window, up to 128K output tokens; 30-day data retention, no zero-retention option
Why It Matters
The grace period ending is an immediate cost event for millions of subscribers and a code-level breaking change for integrations that treat all non-200 responses as failures. Teams migrating from Opus 4.8 also need to rework thinking-block handling since raw reasoning is no longer returned. At current pricing, Fable 5 positions Anthropic to compete directly with GPT-5.5 on cost-per-token at frontier capability.
Read More
- Introducing Claude Fable 5 and Claude Mythos 5 — Anthropic Docs
How this story unfolded
Tools in this story
- LLM
- — Large Language Model의 약자로, '거대 언어 모델'이라고 해요. ChatGPT, Claude 같은 AI가 바로 LLM이에요. 엄청나게 많은 텍스트를 학습해서 사람처럼 글을 쓰고 대화할 수 있어요.
- API
- — Application Programming Interface의 약자예요. 개발자가 AI 기능을 자기 앱에 연결할 때 사용하는 방법이에요. 일반 사용자는 몰라도 되지만, 앱 개발할 때 필요해요.