Claude Fable 5.1 explained: Anthropic 2026 flagship and Mythos
Anthropic’s 1 September release: Fable 5.1 is the public flagship with cheaper cache; Mythos 5.1 is the trusted-access twin. The story is layered safeguards, not an empty leaderboard.
On 1 September 2026 Anthropic released Claude Fable 5.1 and the same-weight Claude Mythos 5.1. Public materials frame a coding and knowledge-work flagship, not another empty leaderboard drop.
The company blog says Fable 5.1 is generally available; Mythos 5.1 is limited to trusted-access programs. The Verge, the same day, reported that Anthropic was answering complaints on price, data retention, and over-strict safeguards.
One weight file, two safeguard stacks
Fable and Mythos share weights. The split is high-risk dual-use work in biology and cybersecurity. The system card lists a June 2026 knowledge cutoff; platform docs list a 1M-token context and 128K max output.
Fable 5.1
Public flagship. It may find software bugs, not write exploits. Pentests and binary scans still go to Opus.
Mythos 5.1
Same capability, looser domain safeguards. Access is mostly US-vetted orgs; the life-sciences program is built with the US government.
Claude Security
Enterprise codebase scanning with human-reviewed patches, now powered by Mythos 5.1.
Price: inputs unchanged, cache cheaper
-
1
Read the list price
$10 / M input and $50 / M output, same as Fable 5. Cache reads drop 75% to $0.25 per million tokens.
-
2
Read Anthropic’s own bill math
On four weeks of August 2026 default-effort usage: about 25% less on typical work; up to about 45% on context-heavy agent jobs.
-
3
Read the storefront
Developers call
claude-fable-5-1. AWS, Google Cloud, and Microsoft Foundry carry it. Mythos is not on the public rate card.
What else lines up on the official table
| Benchmark | Fable 5.1 | Compared with |
|---|---|---|
| Terminal-Bench 4.0 | 55.8% (Mythos 60.9%) | Fable 5 42.0%, Opus 5 52.3% |
| Humanity's Last Exam (no tools) | 60.9% | Fable 5 57.8%, Opus 5 56.6% |
| CursorBench 3.2.0 | 73.4% | Fable 5 70.5%, GPT-5.6 Sol 67.2% |
Anthropic writes that Fable 5.1 and Mythos 5.1 are two configurations of one model; the public build adds limits on high-risk dual-use biology and cyber tasks.
- Enterprise retention
- Enterprise Frontier Safeguards is planned in phases this fall, with data on the customer cloud. Eligible customers can use Fable 5.1 with zero data retention until then.
- EU watermark
- Models released after 2 August 2026 carry an invisible watermark under the EU AI Act code of practice; the detection API is in a limited preview.
When a team needs one picture
After reading, groups usually want one sketch: public flagship, trusted twin, cache bill, Opus redirects. No extra meeting suite required. Open a short space on tidemeet, or see how to create a space. For an eval-isolation incident, read METR’s OpenAI eval investigation; for another lab design, see DeepMind’s double-blind enclave eval.
# map
fable-ga → mythos-trusted → cache-bill → opus-redirectsQuestions
Are Fable 5.1 and Mythos 5.1 two models?
No. The blog and system card both say one weight file and two safeguard levels. Mythos is not self-serve.
Did input and output prices drop together?
No. $10 / $50 stayed put. The savings come mainly from cache reads at $0.25 per million tokens.
Can Fable run a penetration test now?
Anthropic allows vulnerability discovery. Pentests, exploit writing, and binary-based scanning still go to Opus.
What does this have to do with a temporary meeting tool?
Nothing product-tied. If you only need one sketch of the split, temporary meeting tools explains when a short browser space is enough.