Claude Fable 5.2 arrives $2 cheaper per million input tokens than 5.1

Anthropic has added Claude Fable 5.2 to its models overview page, where it appears with the API ID claude-fable-5-2. It costs $8 per million input tokens and $40 per million output tokens, according to the page.

Specs and prices on the models page

The page positions Fable 5.2 for demanding reasoning and long-horizon agentic work. It lists a 1M-token context window, up to 128K output tokens, always-on adaptive thinking, a default effort of high, and a reliable knowledge cutoff of June 2026. Its latency is rated moderate, the same as Claude Opus 5.5.

Anthropic’s guidance is to start with Opus 5.5 for most workloads and move to Fable 5.2 when evaluations at higher effort on Opus still fall short. Opus 5.5 is listed at $4 per million input tokens and $20 per million output, so Fable 5.2 costs twice as much per token. Sonnet 5.5 is $2 and $10, and Haiku 5.5 starts from $0.10 and $0.50.

The page says Fable 5.2 will not be retired sooner than November 9, 2027. Its model ID table shows the model on the Claude API, Claude Platform on AWS, and Claude Models on Google Cloud, both Google managed and Anthropic managed. The Amazon Bedrock and Microsoft Foundry columns show a dash for Fable 5.2, while the other three models are listed there.

A lower price and a faster tier

The company’s November 9 release notes add that Fable 5.1 and Mythos 5.1 were priced at $10 and $50, so the new model is 20 percent cheaper on both input and output. The notes also describe a fast mode for Fable 5.2 on the Claude API, with up to 2.5 times higher output speed at $16 per million input tokens and $80 per million output tokens, double the standard rate. It is switched on with speed: "fast" in the request, without a beta header.

On X, the @claudeai account said it was introducing Fable 5.2 and Claude Mythos 5.2, “designed to tackle the most daunting technical challenges.” According to the release notes, Mythos 5.2 offers the same capabilities as Fable 5.2 but only to organizations verified through Anthropic programs such as the Cyber Verification Program.

Conditions that come with the model

The release notes attach several requirements, and they matter for anyone planning a migration. Thirty-day data retention is required, as with Fable 5.1, and zero data retention is not available unless Anthropic expressly authorizes it. A prefix check applies to every account: if code edits the system prompt, the tools or an earlier message and then sends back a thinking block from a later turn, the request returns a 400 error. Thinking blocks are bound to the account that produced them, or a linked account.

Computer use requires the computer_toolset_20260801 toolset, and the older computer_20251124 tool returns a 400 error. In the beta advisor tool, an executor running Fable 5.2 or Mythos 5.2 accepts only those two models as its advisor, while every executor model can use Fable 5.2 as an advisor.

What the sources leave unanswered

The model is the continuation of a line that began with Claude Fable 5 and Mythos 5 on June 9, 2026, which The Decoder and other outlets covered at the time. Those reports described Mythos 5 as the same underlying model with fewer safeguards, restricted to partners, and said Fable 5 reroutes some sensitive requests to Opus 4.8. The pages read for this article do not say whether Fable 5.2 keeps that routing.

Neither the models page nor the release notes include benchmark results, so the claim that Fable 5.2 improves on coding and other areas rests on Anthropic’s X post alone and is not independently verified. The page also does not say whether Fable 5.2 is available in Claude consumer plans, how fast mode is rate limited, or why Bedrock and Foundry are missing. The release notes do not mention fast mode for Mythos 5.2.

The price cut arrives as Anthropic is marketing an IPO, which aiholics covered in its report on the IPO marketing and falling AI valuations. Nothing in the sources links the two. For developers, the practical change is narrower: the top tier now costs less than before, and the gap to Opus 5.5 is a clean factor of two, which makes the choice a question of whether higher-effort Opus runs still fail on a given task.

Early use cases are already surfacing; an earlier aiholics article described Claude designing three working gadgets with Diode using the new model.