IA que trabaja
Anthropic released Claude Fable 5.2 on November 9, 2026, together with Claude Mythos 5.2. In short: it’s their most powerful model available to everyone, up to 30% faster than Fable 5.1, 20% cheaper via API ($8 per million input tokens and $40 per million output tokens, previously $10 and $50), and with 55% fewer unnecessary blocks. It’s now available in all paid Claude plans, in Claude Code, and via the API.
That said, just because it’s the most powerful doesn’t mean you should use it for everything. Anthropic itself recommends Opus 5.5 as your everyday model and switching to Fable when the task really demands it. Let’s get into the details.
What is Claude Fable 5.2 (and what is Mythos 5.2)?
Fable and Mythos 5.2 are the same model under the hood. The difference is in access:
- Fable 5.2 has extra protections for cybersecurity and biology, so it’s open to anyone with a paid plan or API access.
- Mythos 5.2 is the version without those restrictions, for verified organizations—like cybersecurity or life sciences teams that pass Anthropic’s verification programs.
For a typical business, Fable 5.2 is the one you’ll use.
What’s changed compared to Fable 5.1
| Fable 5.1 | Fable 5.2 | |
|---|---|---|
| API price (input / output, per million tokens) | $10 / $50 | $8 / $40 |
| Speed | Baseline | Up to 30% faster |
| Fast mode | No | Yes, up to 2.5× faster ($16 / $80) |
| Unnecessary blocks | Baseline | 55% fewer overall |
| Context window | 1 million tokens | 1 million tokens input, 128,000 output |
The reduced blocking is especially noticeable if you work with technical topics: Anthropic reports 60% fewer blocks on cybersecurity questions and 40% fewer in biology, while maintaining the same protection against dangerous uses. When something still gets blocked, the answer is provided by another model (Opus), instead of just returning nothing.
Is it better than Opus 5.5?
According to Anthropic’s own benchmarks, yes—and sometimes by a fair margin. Here are the ones most relevant to business work:
| Benchmark | What it measures | Fable 5.2 | Opus 5.5 |
|---|---|---|---|
| FrontierCode 2.0 | Coding as an agent | 49.1% | 45.5% |
| FrontierSWE v2 | Coding as an agent | 69.9% | 62.3% |
| AutomationBench | Automating business processes | 51.4% | 42.5% |
| Remote Labor Index | Real remote work tasks | 41.3% | 21.3% |
Two honest caveats. First, these are Anthropic’s own numbers, not yet third-party verified. Second, it doesn’t win everywhere: for example, in AutomationBench, Gemini 4 Carbon scores 53.9%, a bit higher.
What stands out most to me is the leap in long tasks: the Remote Labor Index nearly doubles compared to Opus 5.5. That’s exactly the kind of work I see in automations—many hours, many steps, and you don’t want it to get lost along the way.
How much does Claude Fable 5.2 cost?
It depends on how you use it:
- In Claude plans (Pro, Max, Team, Enterprise): it’s included. In Pro and standard Team seats, it can use up to 50% of your weekly limit. Also, for the launch, Anthropic is giving extra Fable usage to paying subscribers.
- Via API: $8 per million input tokens and $40 per million output tokens. With batch processing, it’s half: $4 and $20. Cache reads are $0.20 per million, which is where you really save with agents that repeat context.
- Fast mode: double the price ($16 and $80), but generates up to 2.5× faster.
For reference: a task that reads 200,000 tokens and writes 20,000 costs about $2.40 with Fable 5.2 at standard pricing. If that context repeats and is cached, the cost drops a lot.
When is it worth using Fable instead of Opus 5.5?
This is the key question, and the short answer is: when Opus isn’t enough. Anthropic’s documentation says it plainly: start with Opus, and move to Fable if your high-effort tests with Opus still don’t deliver what you need.
I’d use it for:
- Long programming projects with Claude Code: migrations, major refactors, or building a whole app.
- Automations with many steps that need to work right the first time, where a failure costs more than the model.
- Heavy analysis: cross-referencing lots of documentation, contracts, or data and drawing conclusions.
And I’d stick with Opus 5.5—or even Sonnet 5.5 or Haiku 5.5—for everyday tasks: replying to emails, summarizing, classifying, drafting, or customer service chatbots. There, Fable is overkill.
One idea I really like: Fable 5.2 can act as an advisor or orchestrator for smaller models. That is, the bulk of the work is done by a cheaper model, and only consults Fable when it gets stuck. High quality at a lower cost.
How to use Fable 5.2 in Claude Code
If you already use Claude Code, it’s straightforward:
- Update Claude Code to the latest version.
- Choose the model with
/modeland select Fable 5.2. - If you want fast mode, activate it with
/fast. Note: it uses more quota. - Adjust effort based on the task: in the apps and in Claude Code, it defaults to medium. Raise it to high or very high for tough jobs, and lower it for easy ones—that’s where you save on usage.
If you’re starting from scratch, my Claude Code guide covers everything from installation to running it solo, and if you want to connect your own tools, see the MCP guide.
What changes if you use the API?
Switching models is as simple as setting claude-fable-5-2, but there are four changes that might break code that worked with 5.1: restrictions on editing previous messages now apply to all accounts; Fable 5.2 reasoning blocks can only be read by Fable or Mythos 5.2; the old computer tool (computer_20251124) no longer works outside Bedrock; and advisor mode requires an advisor from the same family. Also, adaptive reasoning is always on and you can’t force use of a specific tool. Before switching in production, review the official migration guide.
What does all this mean for a small business?
What was expensive and slow a year ago is now cheaper and faster, and agents can handle longer tasks without losing track. In practice, things like an agent that checks orders, invoices, and emails every morning and only flags the unusual, or keeps your online store up to date, are now more reliable and cost less.
If you want to take advantage of this without wrestling with the tech, I can set up AI agents for you or help you decide what makes sense with an AI consulting session for businesses. And if you’d rather learn to do it yourself, in El Atajo I teach how to work with Claude Code step by step.
