Claude Fable 5.2 costs 20% less than Fable 5.1, which is a good reason to switch. Anthropic’s docs also list four changes that can break your code. If Claude Code, claude.ai, Claude Managed Agents or the Claude Agent SDK builds your requests, nothing changes for you. If your own code builds them, spend the half hour.
1. The prefix check no longer cares how old your account is
The API checks whether anything before a thinking block has changed since the model produced it: the system prompt, the tools, or an earlier message. If your code sends back an invalidated block and prefix_mismatch_behavior is unset, Fable 5.2 and Mythos 5.2 return a 400, whatever the account’s age.
On Fable 5.1 the same error only fires on accounts created on or after August 31, 2026. That is the trap: code that rewrites its history runs clean on an older account today and fails after the switch. The error message reads “The block is bound to a different conversation”.
The escape hatch is the thinking-binding-controls-2026-08-01 beta header with thinking.block_binding.prefix_mismatch_behavior set to "drop_block". The response then lists each dropped block in input_transformations. Anthropic calls this a stopgap rather than a fix, because dropping blocks pushes output tokens up.
You are safe if system and tools stay identical for the whole conversation, your code only appends to messages, and every assistant turn goes back exactly as the API returned it: every block, same order.
2. Only Fable 5.2 and Mythos 5.2 read Fable 5.2 thinking blocks
Every thinking block records which model produced it. Fable 5.2 and Mythos 5.2 read each other’s blocks plus those from Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1, Mythos 5.1, Opus 5, Fable 5, Mythos 5 and earlier Opus, Sonnet and Haiku models. No other model reads theirs.
So moving up keeps the conversation’s reasoning and moving back loses it, including to Opus 5.5 and to Fable 5.2’s fallback models Opus 4.8 and Opus 5. The request does not fail: the API drops the block before the model sees it, and dropped blocks are not billed.
3. computer_20251124 errors out everywhere except Bedrock
Fable 5.1 accepts computer use two ways: as the computer_toolset_20260801 toolset and, with the computer-use-2025-11-24 beta header, as the older computer_20251124 tool. Fable 5.2 and Mythos 5.2 only know the toolset. A request declaring the old tool returns 400 invalid_request_error on every platform except Amazon Bedrock, where it still works as it does on 5.1.
4. The advisor has to move with the executor
With the advisor tool, still in beta, the advisor must be at least as capable as the executor. An executor on Fable 5.2 or Mythos 5.2 accepts only Fable 5.2 or Mythos 5.2 as its advisor. Anything else, Fable 5.1 and Mythos 5.1 included, returns 400 invalid_request_error.
Two more things that throw no error but cost you something
Both models require 30-day data retention and are not available under zero data retention without express authorization from Anthropic. And thinking blocks they produce only work in the account that produced them, or in an account linked to it.
Switching today means checking one line of code
Of the four, the prefix check is the one nobody notices until it blows up. The other three sit in your code as names: you grep for computer_20251124, for the advisor config, for the fallback model. The prefix check you find only by looking at whether your code touches an earlier message, the system prompt or the tool list mid-conversation. Agent loops that reveal and hide tools as they go do exactly that, rather happily.