What Is Claude Fable 5.2? Effort, Fast Mode and the Real Cost per Task
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026: the same underlying model at two safeguard levels, with Fable as the public one.
AI Stella and Liam are AI hosts with AI-generated voices. About our AI hosts
Key takeaways
- Fable 5.2 and Mythos 5.2 are the same underlying model at two safeguard levels: Fable is public, Mythos is for verified organizations. Context is 1,000,000 tokens in, max output 128,000.
- Pricing dropped 20% from 5.1: $8 per million input tokens and $40 output, batch at $4 and $20, cache reads $0.20. Fast mode is up to 2.5x faster at double price, $16 and $80.
- On FrontierCode 2.0, max effort is a trap: 45.7% at $44.70 per task, below xhigh at 49.1% and $14.19. High effort is the sweet spot at 45.3% and $6.72.
- Migration gotchas return 400 errors: adaptive thinking cannot be disabled, forced tool_choice breaks, and temperature, top_p and top_k must be omitted. The model also requires 30 day data retention.
Show notes
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026: the same underlying model at two safeguard levels, with Fable as the public one. In this episode of Agentic Pod, Liam and Stella cover the new $8 and $40 pricing, the 1,000,000 token context, what fast mode really charges you, the 5 effort levels with the real FrontierCode 2.0 cost per task for each one, the migration gotchas that return 400 errors, and the honest verdict on when Fable beats your current daily driver.
What we cover:
- What Fable 5.2 and Mythos 5.2 are: one model, two safeguard levels, Fable public and Mythos for verified organizations
- Pricing: $8 per million input tokens and $40 output, down 20% from Fable 5.1, batch at $4 and $20, cache reads $0.20
- Fast mode: up to 2.5x faster generation at double price, $16 and $80, in Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry
- Effort levels on FrontierCode 2.0: low 34.2% at $2.10 per task, medium 37.6% at $2.48, high 45.3% at $6.72, xhigh 49.1% at $14.19, max 45.7% at $44.70
- Migration gotchas: adaptive thinking cannot be disabled, forced tool_choice returns 400, temperature and top_p and top_k return 400, 30 day data retention required
- A real workflow: keep your current model as the daily driver and route only the longest, hardest runs to Fable at high effort
Chapters
Questions answered in this episode
What is Claude Fable 5.2?
Anthropic's public model released November 9, 2026, the same underlying model as Mythos 5.2 at a public safeguard level, with a 1,000,000 token context and 128,000 max output. The model ID is claude-fable-5-2.
How much does Claude Fable 5.2 cost?
$8 per million input tokens and $40 output, batch at $4 and $20, cache reads $0.20. Fast mode doubles it to $16 and $80 for up to 2.5x faster generation.
Which effort level should I use?
High for hard work: 45.3% on FrontierCode 2.0 at $6.72 per task. Max scored lower than xhigh (45.7% vs 49.1%) at $44.70 per task, so skip max until your own numbers prove otherwise.
Links & resources
Full transcript
Intro
Hey hey hey, welcome to Agentic Pod! I'm Liam.
And I'm Stella! Liam, Anthropic shipped Claude Fable 5.2 on November 9th, and the top effort level scores worse than the one below it?
Worse, and it costs $44.70 a task against $14.19. That is real, from the FrontierCode 2.0 numbers.
So the max button is a trap?
Sometimes, yeah. By the end you will know the price, the 5 effort levels, fast mode, and when Fable is worth it.
What Fable 5.2 and Mythos 5.2 actually are
Okay, basics first. What did Anthropic actually release?
Two models on November 9th, 2026. Claude Fable 5.2, and Claude Mythos 5.2.
Two names, so two different models?
Same underlying model, I mean, the exact same brain. The difference is the safeguard level.
So Fable is the one I can use?
Right. Fable is the public one. Mythos is only for verified organizations.
Context is 1,000,000 tokens, max output is 128,000. The model ID is claude-fable-5-2.
A million tokens in. A whole code base, plus your bad decisions from March.
Especially those. It is up to 30% faster than 5.1, and its classifiers block about 55% fewer requests.
Fewer false blocks is a real upgrade. Okay, what does it cost?
The price, the speed, and fast mode
Base price is $8 per million input tokens, and $40 per million output tokens.
And that is cheaper than the last one?
Down 20% from 5.1, which was $10 and $50. Batch halves it, to $4 and $20. Cache reads are $0.20.
Okay. Now explain fast mode, because everyone will flip that switch.
Fast mode makes generation up to 2.5x faster. Same model, same answers, just a higher serving rate.
There is always a catch. Price?
Double. $16 input and $80 output per million tokens.
Double the price for speed. That is a taxi meter with a turbo button.
That is exactly what it is. Fast mode lives in Claude Code, the Claude Platform, Google Cloud, and Microsoft Foundry.
So speed is a choice, not a new model. Good. Now the part I actually care about. Effort.
Effort levels and the real cost per task
Five effort levels, right? Walk me through them.
Low, medium, high, xhigh, and max. Effort is how hard the model thinks before it answers.
And the defaults will trip people up?
They will. The API defaults to high. Claude Code and the Claude apps default to medium.
Same model, different default brainpower depending on the door you walk in. Sneaky.
Now FrontierCode 2.0, 132 pull request tasks. Low scores 34.2% at $2.10 per task. Medium scores 37.6% at $2.48.
Cheap so far. Keep going.
High scores 45.3% at $6.72. Xhigh scores 49.1% at $14.19.
And max, the one from the intro?
Max scores 45.7% at $44.70. Lower than xhigh, and more than 3 times the price.
So max is slower thinking that does worse and bills more. My favorite kind of upgrade.
Be nice. One more anchor. High effort roughly matches Opus 5.5 at max effort, for about half the cost per task.
That is the headline for me. But before anyone migrates, what breaks?
Migration gotchas and the honest verdict
3 big ones. First, adaptive thinking cannot be disabled on this model. It thinks adaptively, period.
Okay, what is number 2?
Forced tool choice returns a 400 error. If your code forces tool_choice to any or tool, it breaks. And temperature, top_p, and top_k also return 400. You have to omit them.
So old API code with sampling settings just falls over?
It does. Number 3 is data. The model requires 30 day retention. No zero data retention, unless Anthropic authorizes it.
And Claude Code needs an update, right?
Version 2.1.321 or later. Fable is on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and paid Claude apps.
Quick note for paid plan people, by the way. There was a launch promo week, November 9th to 16th. After that, Fable draws on usage credits on Pro.
Good catch. Now the workflow I would actually run. Keep your current model as the daily driver.
And Fable is the specialist you call in?
Exactly. Route only your longest, hardest runs to Fable at high effort. Measure cost per finished task, not per token.
And xhigh only where you measured a real gain. Not by vibes.
Not by vibes. And Anthropic itself says start with Opus 5.5 for most coding and knowledge work. Move up to Fable only as needed.
The maker telling you not to buy the big one. I respect that.
My verdict: Fable 5.2 at high effort is the sweet spot. Max is a trap until your own numbers prove otherwise.
Same. The step by step guide is in the description. Would you pay double for fast mode, or is high effort enough? Tell us in the comments. See you next time!