Claude Fable 5.2 pricing is $8 in and $40 out per million tokens, 20% below Fable 5.1. Move that traffic after replaying a day of real requests, because three API changes return 400 errors and a fourth drops reasoning. Replay it on Opus too: it costs half as much.
What does Fable 5.2 cost against Fable 5.1?
Anthropic's pricing page lists every line exactly 20% lower on 5.2, cache included.
| Line (per MTok) | Fable 5.1 | Fable 5.2 |
|---|---|---|
| Input | $10 | $8 |
| Output | $50 | $40 |
| 5-minute cache write | $12.50 | $10 |
| 1-hour cache write | $20 | $16 |
| Cache read | $0.25 | $0.20 |
If the work can wait, batch takes 5.2 down to $4 in and $20 out. Fast mode (speed: "fast") buys up to 2.5x the output speed at $16 and $80, twice the standard rate.
The cut doesn't make Fable the cheap option. Anthropic's migration guide prices Opus 5.5 at $4 in and $20 out, half of Fable 5.2. Its what's-new page advises starting with Opus and moving to Fable 5.2 only when your evals on Opus at higher effort still don't meet your requirements, so the decision rests on your own evals. Our Opus 5.5 write-up covers that model's pricing and its own four breaking changes.
Fable 5.2 runs on the Claude API, Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry. If your spend already sits on a cloud commitment, read which contract should carry your Claude spend before you pick an endpoint.
What does the cut do to a real monthly bill?
Assume an internal coding agent that bills 60M uncached input tokens, 20M 5-minute cache-write tokens, 400M cache-read tokens and 25M output tokens a month.
| Line | Fable 5.1 | Fable 5.2 |
|---|---|---|
| Input | $600 | $480 |
| Cache writes | $250 | $200 |
| Cache reads | $100 | $80 |
| Output | $1,250 | $1,000 |
| Total | $2,200 | $1,760 |
You save $440 a month. Output is more than half the bill, so output is where the saving can leak.
Two things can push output up. The API's default effort is high, and Anthropic reports that a line like "answer directly without deliberating" didn't stop 5.2 from thinking at medium and above. The other is the prefix-check stopgap in the next section, which drops mismatched thinking blocks, and Anthropic says dropping blocks can raise output tokens. Suppose those effects add 20% to output, an assumption. Output becomes 30M tokens, or $1,200, the total becomes $1,960, and your saving shrinks to $240.
Anthropic's pages don't give a figure for how much effort moves billed tokens. Replay a day of your requests on claude-fable-5-2 with effort fixed and compare billed tokens against 5.1. Subagents are the other place to look: Anthropic says they can account for a large part of what a session costs.
What breaks if you only change the model ID?
Four things break or change behaviour, and three of them return 400s.
- Edited history: 5.2 checks that nothing before a thinking block has changed since the model produced it, and a mismatch returns "The block is bound to a different conversation". 5.1 only enforces this on accounts created on or after 31 August 2026, so code can pass on an older 5.1 account and fail after the switch. Your engineers' stopgap is a beta header (
thinking-binding-controls-2026-08-01) plus settingprefix_mismatch_behaviortodrop_block, which discards the mismatched block instead of failing. - Computer use: a request declaring the
computer_20251124tool returns a 400 on every platform except Bedrock. Usecomputer_toolset_20260801. - Advisor tool: a 5.2 executor accepts only Fable 5.2 or Mythos 5.2 as advisor. A 5.1 advisor returns a 400, so set the advisor model to
claude-fable-5-2. - Model routers: moving a conversation from 5.2 to any other model except Mythos 5.2 drops its thinking blocks. The request still succeeds, and dropped blocks aren't billed, but the reasoning is gone.
The prefix-check change doesn't touch requests built by Claude Code, claude.ai, Claude Managed Agents or the Claude Agent SDK. For your own code, test on 5.2 itself, since an older 5.1 account hides it.
Which of your people get Fable 5.2 without asking?
Claude Code
Version 2.1.321 makes 5.2 the default Fable model and moves saved Fable 5.1 picks to fable. On Pro and Max that happens at once. On Team and Enterprise it happens once your organization allows it. Anyone who wants 5.1 back types /model claude-fable-5-1. If you run Team or Enterprise, find out who in your org holds that switch now. Our piece on managed settings that restrict which models Claude Code may use shows where those controls live.
GitHub Copilot
GitHub's changelog says Fable 5.2 is available to Copilot Pro+, Max, Business and Enterprise users. On Business and Enterprise an admin must enable the Claude Fable 5.2 policy in Copilot settings, and it's off by default. It's billed at provider list pricing under usage-based billing, which makes the policy a budget decision, not a toggle. See how Copilot Business and Enterprise billing changed for that context.
Can you take the 30-day data retention?
Both Fable 5.2 and Mythos 5.2 require 30-day data retention and aren't available under zero data retention unless Anthropic expressly authorizes it. Copilot requires retention by default, and enabling its policy is the acknowledgment.
In Copilot, eligible enterprises keep zero data retention under a time-limited exemption through the end of 2026; after that Anthropic's Enterprise Frontier Safeguards are required.
So the question for your security owner is simple. Does any team here depend on a zero-retention agreement? If yes, keep it off 5.2 until someone gets written authorization. If no, retention is a sign-off, not a project. How data retention works for verified orgs covers the verification route.
What this means for your AI tooling budget
Our take: treat this as a bill-and-test change, not an upgrade project.
Switch your Fable 5.1 traffic once the checks pass on a replayed day of real requests, with effort set explicitly, and run Opus 5.5 through that same replay. If Opus meets your bar, it costs half as much as Fable 5.2 and the cheaper line item is the decision. Fable 5.1 stays listed on Anthropic's pricing page, so you choose the date.
Then settle two admin defaults in one meeting. Who in Team and Enterprise gets the Claude Code move, and does the Copilot policy go on, given that it bills at list price and requires retention? Whoever owns the budget and whoever owns security should both be in that room.
For other model and pricing changes that hit a budget, see our AI news hub.
FAQ
How much does Claude Fable 5.2 cost?
$8 per million input tokens and $40 per million output tokens, 20% below Fable 5.1's $10 and $50. Cache writes are $10 (5 minutes) and $16 (1 hour), and cache reads are $0.20. Batch is $4 and $20. Fast mode is $16 and $80.
Do I have to move off Claude Fable 5.1?
No. Anthropic's pricing page still lists Fable 5.1 and gives no retirement date. At the 20% gap, though, every month on 5.1 costs you the difference unless your tests show 5.2 raises your output tokens.
Will Claude Fable 5.2 work with my Fable 5.1 code?
Not necessarily. Edited thinking history, computer_20251124, a 5.1 advisor on a 5.2 executor and model routers all need checking. The prefix check needs no work if Claude Code, claude.ai, Managed Agents or the Agent SDK build your requests.
Is Claude Fable 5.2 on in GitHub Copilot by default?
Not for Business and Enterprise. An admin has to enable the Claude Fable 5.2 policy in Copilot settings, and it's off by default. It's billed at provider list pricing under usage-based billing and requires data retention.
Does Claude Fable 5.2 work with zero data retention?
Not unless Anthropic expressly authorizes it. Fable 5.2 and Mythos 5.2 both require 30-day data retention. In Copilot, eligible enterprises keep zero retention under a time-limited exemption through the end of 2026, then need Anthropic's Enterprise Frontier Safeguards. If a team depends on zero retention, keep it off 5.2 until you have that authorization in writing.