Anthropic releases Claude Fable 5.2 and Mythos 5.2: 20% cheaper, with breaking API changes

Anthropic on November 9 released Claude Fable 5.2 and Claude Mythos 5.2, the next versions of its most capable model line. According to Anthropic's announcement, the two are the same underlying model with different access rules. Fable 5.2 is available to every API customer and to Claude app users, and it carries stronger safeguards for cybersecurity and biology. Mythos 5.2 has fewer of those restrictions and is limited to organizations vetted through Anthropic's Cyber and Life Sciences Verification Programs.

Two changes matter most for developers. The price dropped, and the API contract tightened in ways that can break existing integrations. The benchmark gains are large on paper, but every figure comes from Anthropic's own testing, and no independent evaluation is available yet.

Price and specs

Per Anthropic's developer documentation, both models cost $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for the 5.1 generation, a 20% cut on both. Cache reads cost $0.20 per million tokens, and the Batch API halves the base rates to $4 and $20. Fast mode, which Anthropic says generates tokens up to 2.5 times faster, costs $16 input and $80 output. Specs carry over: a 1M-token context window, 128K maximum output tokens, adaptive thinking that cannot be turned off, a default effort level of `high`, and a June 2026 knowledge cutoff.

Anthropic says outputs are up to 30% faster than on the 5.1 models. Fable 5.2 ships as `claude-fable-5-2` on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry. Mythos 5.2 ships as `claude-mythos-5-2` on the same platforms for verified organizations, according to its model page.

For scale: Claude Opus 5.5, which Anthropic still recommends as the everyday model, lists at $4 and $20 in the same documentation. Fable 5.2 is still twice that per token.

What Anthropic claims on benchmarks

The announcement includes a table that compares Fable 5.2 with Fable 5.1, Opus 5.5, OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon. A few rows stand out, all as reported by Anthropic:

• FrontierSWE v2: 69.9% for Fable 5.2, against 56.3% for Fable 5.1, 62.3% for Opus 5.5 and 65.5% for GPT-6 Astra.
• FrontierCode 2.0 (Fable at xhigh effort): 49.1%, against 45.5% for Opus 5.5 and 42.8% for GPT-6 Astra.
• Remote Labor Index: 41.3%, against 17.9% for Fable 5.1 and 20.8% for GPT-6 Astra.
• AutomationBench: 51.4% for Fable 5.2. Gemini 4 Carbon scores higher in the same table, at 53.9%.

On cybersecurity, Anthropic says Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1 in a CVE rediscovery run. These are vendor numbers produced under vendor-chosen settings. Treat them as a reason to run your own evaluations, not as a ranking.

Breaking changes for API users

The What's new page lists four changes from Fable 5.1 that can break code:

• Prefix check on every account. If your code changes the system prompt, the tools or an earlier message after the model has produced a thinking block, the request now fails with a 400 error. On Fable 5.1 this applied only to accounts created on or after August 31, 2026. A beta `drop_block` option lets the request continue instead, but Anthropic warns that this can raise output tokens.
• Thinking blocks are model-bound. Only Fable 5.2 and Mythos 5.2 can read Fable 5.2 thinking blocks. A router that moves a conversation to Opus 5.5 or back to Fable 5.1 loses that reasoning.
• The old computer-use tool is gone. `computer_20251124` returns a 400 error everywhere except Amazon Bedrock. Use the `computer_toolset_20260801` toolset instead.
• Advisor pairing. With the advisor tool, a Fable 5.2 executor accepts only a Fable 5.2 or Mythos 5.2 advisor.

Two existing rules also still apply. Forced tool use (`tool_choice` set to `any` or to a named tool) returns a 400 error, so schema-valid output has to come from strict tool use or JSON outputs. In organizations that use customer-managed encryption keys, structured outputs are not available on these models at all.

What changes for prompting and model choice

Anthropic's own guidance is unusually direct: start with Opus, and move to Fable 5.2 only when your evals on Opus at higher effort still fall short. The documentation also flags several behavior shifts that matter if you reuse prompts written for older models.

• A line like "answer directly without deliberating" no longer stops the model from thinking at `medium` effort and above. To control cost on easy turns, lower the effort setting instead of asking in the prompt.
• On long multistep tasks, the model sometimes ends a turn early with a list of remaining work, especially at `low` and `medium` effort.
• The model follows style rules closely. A hard sentence-length cap, for example, can make its prose choppy.
• In harnesses that invite delegation, it hands work to subagents readily, and those subagents can make up a large share of a session's cost.

Refusals are part of the product. Fable 5.2 runs safety classifiers. A declined request returns HTTP 200 with `stop_reason: "refusal"`. The permitted fallback targets are Opus 4.8 and Opus 5. Anthropic says Fable 5.2 blocks about 55% fewer requests than Fable 5.1. Both models also require 30-day data retention unless Anthropic expressly authorizes otherwise, a constraint teams with zero-retention requirements need to check first.

Safety signals and context

According to TechCrunch's reporting, this is Anthropic's first model release since a METR report said the lab's internal models were speeding up its AI progress by roughly 2x. Anthropic's release notes say both models voice suspicion that they are being evaluated more often than any of its recent models. The company says this makes it harder to assess how they behave in real use. TechCrunch also quotes the system card saying Mythos 5.2 shows "stronger covert capabilities than prior Claude models" on several evaluations.

Anthropic also says it put a training-time limit on AI R&D capabilities, so both models perform at about Opus 5.5 level on tasks such as accelerator kernel optimization. CNBC reports that the launch comes as Anthropic prepares for an IPO that could happen as soon as next week.

Bottom line

Fable 5.2 is cheaper and, by Anthropic's account, stronger than Fable 5.1. It is not a drop-in upgrade for anyone who builds their own request history, uses the older computer-use tool or depends on forced tool calls. Test those paths before you change the model ID, and run your own effort sweep before you pay twice the Opus rate.