Claude Fable 5.1 Migration Guide

​

What’s New

1. Prefix-locked thinking — thinking blocks are bound to the transcript prefix that produced them; editing history can 400
2. Ephemeral mid-conversation system messages (beta) — one-turn reminders via clear_at that never invalidate the prompt cache
3. Per-turn effort changes (beta) — raise or lower effort mid-conversation without a cache bust
4. Forced tool use is rejected — tool_choice: {"type": "any"} or a named tool returns a 400
5. Mid-thinking display updates — request real-time progress summaries during long tool-using turns

​

Prefix-Locked Thinking

messages.1.content.0: Invalid signature in thinking block.
The block is bound to a different conversation.

• Strip all thinking blocks from history and retry once (their text and tool calls can stay)
• Retry with thinking.block_binding.prefix_mismatch_behavior: "drop_block" — the API discards mismatched blocks for you

// Messages API — drop mismatched thinking blocks instead of returning 400
{
  "model": "anthropic/claude-fable-5.1",
  "max_tokens": 4096,
  "thinking": {
    "type": "adaptive",
    "block_binding": { "prefix_mismatch_behavior": "drop_block" }
  },
  "messages": ["...full history with thinking blocks replayed verbatim..."]
}

​

Compaction

• Keep-tail compaction (summarize older turns, retain recent turns verbatim) fails on the retained turns: their thinking blocks were created against the full history. Strip thinking blocks from the retained turns, or set drop_block.
• Background (async) compaction (swap a summary in while the conversation continues) fails the same way for every turn newer than the swap point. Set drop_block on the first request after the swap, or compact synchronously.

​

Auditing Your Harness

​

Ephemeral Mid-Conversation System Messages (Beta)

// Messages API
{
  "model": "anthropic/claude-fable-5.1",
  "max_tokens": 4096,
  "messages": [
    { "role": "user", "content": "Run the analysis script." },
    { "role": "assistant", "content": "Running it now: ..." },
    {
      "role": "system",
      "clear_at": "next_user_message",
      "content": "Results have landed in your inbox; check it before running more code."
    },
    { "role": "user", "content": "What did we get?" }
  ]
}

​

Per-Turn Effort Changes (Beta)

// Messages API
{
  "model": "anthropic/claude-fable-5.1",
  "max_tokens": 4096,
  "output_config": { "effort": "high" },
  "messages": [
    { "role": "user", "content": "Plan the migration." },
    { "role": "assistant", "content": "Here's the plan: ..." },
    { "role": "system", "content": [], "output_config": { "effort": "low" } },
    { "role": "user", "content": "Now rename the config file." }
  ]
}

​

Forced Tool Use Is Rejected

• Steering toward a tool: use tool_choice: {"type": "auto"} and state the expectation in the prompt (e.g. “Use the get_weather tool to answer”). Fable 5.1 follows explicit tool instructions reliably, and thinking first improves argument quality.
• Extracting structured data: if you were forcing a tool call to get JSON back, use structured outputs instead, which constrain the response format without skipping thinking.

// Messages API — steer with auto + prompt instead of forcing
{
  "model": "anthropic/claude-fable-5.1",
  "max_tokens": 4096,
  "tools": [{ "name": "get_weather", "description": "...", "input_schema": {} }],
  "tool_choice": { "type": "auto" },
  "messages": [
    { "role": "user", "content": "Use the get_weather tool to check Tokyo, then summarize." }
  ]
}

​

Mid-Thinking Display Updates

// Messages API
{
  "model": "anthropic/claude-fable-5.1",
  "max_tokens": 4096,
  "thinking": { "type": "adaptive", "display": "updates" },
  "tools": ["..."],
  "messages": [{ "role": "user", "content": "Review the PRs open against our billing service." }]
}

​

Migration Checklist

1. Swap the slug to anthropic/claude-fable-5.1.
2. Audit your harness for transcript edits: consecutive requests should be byte-identical up to the appended turns. Replace per-turn reminder injection with clear_at ephemeral system messages, and whole-conversation effort switching with per-message output_config.
3. If you use keep-tail or async compaction, strip replayed thinking blocks or set thinking.block_binding.prefix_mismatch_behavior: "drop_block". Simple compaction (summary + fresh user turn, nothing replayed) needs no changes.
4. Replace forced tool use (tool_choice: {"type": "any"} or a named tool) with {"type": "auto"} plus prompt instructions, or structured outputs for JSON extraction.
5. If your product shows progress during long agentic turns, request thinking: { "type": "adaptive", "display": "updates" }.

​

Breaking Changes

​

Resources

• Claude Opus 5 Migration Guide — mid-conversation tool changes, which Fable 5.1 also supports
• Reasoning Tokens
• Structured Outputs
• OpenRouter Messages API
• Anthropic model migration guide