Anthropic launches Claude Fable 5.2 and Mythos 5.2
Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9: the same underlying model with different safeguards, faster output and a lower $8/$40 price. What improved, the specs, how to switch in the API and Claude Code, and when Opus 5.5 is still the better pick.
Updated November 20, 2026: Expanded with specs, access rules, migration changes, Claude Code setup and cost-per-task figures.
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026. Anthropic pitches them for its most ambitious, open-ended technical work, beyond what it recommends Opus 5.5 for. It reports gains in coding, reasoning and vision, says both models generate output up to 30% faster than the 5.1 generation, and cuts the price to $8 per million input tokens and $40 per million output tokens.
One model, two ways in
Fable and Mythos are the same underlying model; what differs is the safeguards and who can use them. Fable ships with classifiers that block high-risk cybersecurity work, such as exploit generation, and most research-biology tasks. A blocked request is answered by Opus 4.8 (cyber) or Opus 5 (biology) instead. Anthropic counted about 55% fewer blocks than with Fable 5.1 in early testing.
Mythos 5.2 is the same model without certain of those safeguards. Only organisations admitted to Anthropic's Cyber or Life Sciences Verification Program can use it. Claude Enterprise customers can also use Claude Security, Anthropic's code vulnerability scanner, which draws on models including Mythos 5.2; that is a scanning product, not direct Mythos access.
What Anthropic says improved
All of the following are Anthropic's own numbers, not independent tests, with safeguards on and at each model's highest available effort unless noted. They are a selection from a longer table:
- FrontierCode 2.0 (would maintainers merge the change?): 49.1% at Xhigh effort, against 45.5% for Opus 5.5.
- FrontierSWE v2: 69.9%, against 62.3% for Opus 5.5.
- Remote Labor Index: 41.3% at High effort, against 21.3% for Opus 5.5.
- CVE Rediscovery (Mythos 5.2, 1,829 published CVEs): 89.6%.
- Cost per task: on FrontierCode 2.0, Fable 5.2 at High effort scored about level with Opus 5.5 at Max for roughly half the cost per task. On Perplexity's WANDR benchmark, at Medium effort it matched Opus 5.5's top score for about 40% of the cost.
Anthropic also notes a deliberate limit: in accelerator-kernel optimisation and LLM-architecture research, the model performs at about Opus 5.5's level.
Specs and availability
- Model IDs: claude-fable-5-2 and claude-mythos-5-2 on the Claude API; global.anthropic.claude-fable-5-2 on Amazon Bedrock.
- 1M-token context, 128K maximum output, adaptive thinking always on. Anthropic's models table gives Fable 5.2 a June 2026 reliable knowledge cutoff and no retirement before November 9, 2027.
- Fable 5.2 runs in the Claude apps, the Claude API, Amazon Bedrock, Claude Platform on AWS, Claude Models on Google Cloud and Microsoft Foundry.
- Price: $8 input and $40 output per million tokens, down from $10 and $50. Fast mode (up to 2.5 times higher output speed) costs $16 and $80, is not on every platform, and for Mythos is limited to Verification Program members.
- Both need 30-day data retention unless Anthropic authorises zero retention. Enterprise Frontier Safeguards (private beta) runs safety monitoring on data kept in the customer's own infrastructure; eligible customers can stay on zero retention while setting it up.
How to switch
On the API, the model ID changes to claude-fable-5-2. The API default effort for Fable 5.2 is high; the Claude apps and Claude Code default to medium. A minimal request, adapted from Anthropic's own docs example, looks like this:
import anthropic
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-fable-5-2",
max_tokens=4096,
messages=[
{"role": "user", "content": "Refactor this module to use dependency injection"}
],
)The model string is not always the whole job. The release notes list changes that can break existing code:
- Fable 5.2 enforces the preserved-thinking prefix check on every account. Suppose your code edits the system prompt, the tools or an earlier message, then sends back a thinking block from a later turn. The request returns a 400 error.
- Computer use needs the computer_toolset_20260801 toolset. The older computer_20251124 tool returns a 400 error everywhere except Amazon Bedrock.
- With the advisor tool (beta), a Fable 5.2 or Mythos 5.2 executor accepts only Fable 5.2 or Mythos 5.2 as its advisor.
- Thinking blocks travel one way. Fable 5.2 reads other models' thinking blocks, but no other model reads Fable 5.2's; when a conversation moves to another model, the API drops them.
In Claude Code, no Fable model is the default on any plan, so you have to select it. That needs Claude Code v2.1.321 or later. Inside a session run /model fable, or start a new one like this (in Claude apps gateway sessions the fable alias points to Fable 5, so use /model claude-fable-5-2 there):
claude update
claude --model fableClaude Code starts Fable 5.2 at medium effort. Depending on plan and seat, Fable can bill to usage credits instead of your included limits; the /model picker then marks the row "Requires usage credits". Classifier-flagged requests are re-run on Opus 4.8 or Opus 5, and the transcript notes the switch.
When to use it, and when not
Anthropic does not suggest Fable as a default. Its guidance is to start with Opus 5.5 for most coding and knowledge work, move up to Fable or Mythos for ambitious, open-ended technical problems, or down to Sonnet 5.5 or Haiku 5.5 when those are enough. Dianne Penn, Anthropic's head of product management for research, told CNBC that Fable 5.2 is mainly for "power users or power use cases".
Our take
Fable 5.2 is cheaper and faster than the model it replaces, yet Anthropic still points most people to Opus 5.5, which costs half as much per token. We think that advice is right, with one caveat: Anthropic's own per-task figures suggest Fable at lower effort can undercut Opus at Max. Escalate to Fable 5.2 when Opus 5.5 at higher effort falls short, and measure cost per task on your own workload before moving a pipeline.
Source: Claude Fable 5.2 and Mythos 5.2 (Anthropic), Claude Platform release notes, Models overview (Claude Platform Docs), Model configuration (Claude Code Docs), Fast mode (Claude Platform Docs), Anthropic releases first frontier models since Amodei's call for slowdown as IPO buzz builds (CNBC)
This story first appeared in the claudefox daily briefing for November 10, 2026.