Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026. They are the same underlying model with different doors: Fable ships with cybersecurity and biology safeguards and is available to all users, while Mythos access is granted through Anthropic’s Life Sciences and Cyber Verification Programs.
The interesting part for anyone paying an API bill is not the benchmark talk. It is the price, the speed, and the fact that Anthropic’s own advice is to keep using a different model for everyday work.
What is confirmed about the models
- Shared model: Fable 5.2 and Mythos 5.2 are the same underlying model, separated by safety posture and access.
- Speed: Anthropic says outputs are up to 30 percent faster than Fable 5.1 and Mythos 5.1.
- Standard price: $8 per million input tokens and $40 per million output tokens.
- Fast mode: an optional setting that speeds generation by up to 2.5 times, priced at $16 per million input and $80 per million output.
- Safeguards: Anthropic says Fable’s cyber and biology safeguards are as robust as 5.1’s but produce fewer unnecessary blocks.
Anthropic also lists prompt caching at $0.20 per million tokens to read and $10 per million to write for Fable 5.2, and says the US-only inference option carries a 1.1 times price for input and output tokens.
The price ladder, in one place
Anthropic’s public API pricing page puts the gap in plain numbers, all per million tokens:
- Fable 5.2: $8 input, $40 output.
- Opus 5.5: $4 input, $20 output.
- Sonnet 5.5: $2 input, $10 output.
- Haiku 5.5: $0.10 input and $0.50 output for prompts up to 100,000 tokens, rising to $0.50 and $2.50 above that.
So Fable costs twice what Opus costs, and four times what Sonnet costs, on both input and output. For a workload that generates a million output tokens, that is the difference between $40 and $10 for the same volume of text.
When fast mode is worth double
Fast mode doubles the price per token, which sounds like a bad deal until you do the arithmetic on time. If generation is 2.5 times faster and each token costs twice as much, then the cost of running the model for an hour of wall-clock generation is about 20 percent lower than standard mode, because you finish in 40 percent of the time.
That only helps if wall-clock time is the constraint: an interactive agent, a long chain of tool calls, a build that waits on a model response. If you are batch-processing overnight, fast mode buys nothing but a larger bill, and Anthropic’s 50 percent batch discount runs in the opposite direction.
Who should actually use it
Anthropic’s own guidance is blunt: for most coding and knowledge tasks, start with Claude Opus 5.5 as the daily driver, then move up to Fable or Mythos for the hardest open-ended technical work, or down to Sonnet 5.5 or Haiku 5.5 for simpler tasks. Fable 5.2 can also act as an advisor or orchestrator that other models consult.
That maps onto a practical rule. If your task has a clear specification and a checkable answer, a cheaper model with a good harness will usually get there. If the task is genuinely open-ended, the kind where you cannot write the acceptance test in advance, the expensive model is the one that can change the shape of the answer.
On the consumer side, the tiers have shifted too. Anthropic’s pricing page lists Pro at $20 a month, with Fable available through usage credits; Max at $100 a month with Fable included up to half the weekly limit and $100 in monthly API credits; and Ultra at $200 a month with no five-hour limit for verified accounts, ten times Pro’s weekly usage and $200 in API credits. If you are comparing that with what you already pay, our breakdown of what changed in Claude Opus 5.5 and what it costs is the closer comparison for everyday work.
The claims to treat as vendor claims
Anthropic says that during prerelease testing the model helped its engineers design three working electronic devices from scratch, solve two open problems in the physics of black holes, make progress on robotics evaluations, and formalise the proof of Fermat’s Last Theorem in 30 hours. Those are the vendor’s own descriptions of internal testing. They have not been independently replicated, the tasks are not described in enough detail to reproduce, and a benchmark of “open problems solved” is not a general measure of usefulness for your work.
The same caution applies to “up to 30 percent faster”. That is a manufacturer figure, presumably measured on Anthropic’s own workloads, and it will not transfer exactly to yours.
What we did not do
We did not run a single prompt against Fable 5.2 or Mythos 5.2, and we have no benchmark numbers of our own. We do not have Mythos access, which requires acceptance into a verification program. Prices, model relationships, speed claims and the Opus recommendation all come from Anthropic’s announcement and pricing page, both linked below. No third-party platform availability is claimed here, because Anthropic did not state it in the sources we read.
