WHAT SHIPPED
● Fable 5.1 GA on 1 September 2026, at $10 in and $50 out per million — the same as Fable 5.
● Cache reads: down to $0.25 per million.
● Terminal-Bench-Science: 24.7 to 52.6, self-reported.
● Mythos 5.1 ships as the trusted-access twin, same underlying model with different safeguards.
Same price, roughly double the score
Fable 5.1 is generally available from 1 September at $10 per million input and $50 output, identical to Fable 5. Cache reads drop to $0.25 per million, which matters more than it looks for anyone running repeated prompts against a large fixed context.
The headline figure is Terminal-Bench-Science moving from 24.7 to 52.6. That is a self-reported number on a scientific reasoning benchmark, and it is roughly a doubling.
TREAT THE BENCHMARK AS VENDOR-RUN
A doubling on any benchmark is a large claim, and this one comes from the company selling the model. That is normal and it is not the same as independent replication.
The useful check is whether the gain shows up in your own work. A benchmark that doubles and a workload that does not improve is a common outcome.
The two-tier structure
Mythos 5.1 and Fable 5.1 share the same underlying model. The difference is safeguards — Fable carries additional measures around biology, cybersecurity and machine learning research, and is the version generally available. Mythos is the trusted-access variant.
That split is worth understanding if you are choosing between them, because it is a governance distinction rather than a capability one. You are not picking a better model; you are picking which set of restrictions applies.
Where it sits on price
| Model |
Input / 1M |
Output / 1M |
| Claude Fable 5.1 |
$10 |
$50 |
| Claude Opus 5 |
$5 |
$25 |
| Claude Sonnet 5 |
$3 from 31 August |
$15 |
| GPT-5.6 Sol |
$4 promotional, $5 standard |
$20 promotional, $30 standard |
Fable sits well above the rest of the Claude line, which is the point — it is a different tier rather than a Sonnet alternative. For most work Sonnet 5 or Opus 5 is the right call, and Fable earns its price only on tasks where the capability gap is doing real work.
The cache read detail
At $0.25 per million, cache reads are 40 times cheaper than fresh input. If your workload sends a large fixed context repeatedly — a codebase, a document set, a long system prompt — the effective cost per call is nothing like the headline rate.
That is the number to model against, and it is the one most people skip when comparing rate cards.
Some recent history worth knowing
Fable 5 and Mythos 5 first shipped on 9 June 2026. Access was suspended on 12 June to comply with US Department of Commerce export controls, and restored on 1 July after those controls were lifted. That is a twenty-day gap in availability within the first month of a model's life.
Not a reason to avoid the 5.1 line, but a reason to keep a fallback configured if you build on it. Availability has been interrupted before for reasons unrelated to the technology.
Sources
FAQ
What is Claude Fable 5.1?
Anthropic's Mythos-tier model, generally available from 1 September 2026 at $10 and $50 per million. It shares an underlying model with Mythos 5.1 and carries additional safeguards around biology, cybersecurity and machine learning research.
Did the price change?
No. It matches Fable 5 at $10 in and $50 out, with cache reads dropping to $0.25 per million.
How much better is it?
Self-reported Terminal-Bench-Science went from 24.7 to 52.6. That is a vendor benchmark and has not been independently replicated, so test it on your own workload before assuming the gain transfers.
What is the difference from Mythos 5.1?
The same underlying model with different safeguards. Fable is the generally available version; Mythos is the trusted-access twin. It is a governance distinction rather than a capability one.
Should I use it instead of Sonnet 5?
For most work, no. Fable sits at $10 and $50 against Sonnet 5 at $3 and $15. It earns the difference only where the capability gap does real work for you.
Why do cache reads matter?
At $0.25 per million they are 40 times cheaper than fresh input. If you repeatedly send the same large context, that figure determines your real cost far more than the headline rate.