What Anthropic shipped on September 1
Anthropic made Claude Fable 5.1 generally available on September 1, 2026, shipping it through its own API and through the AWS, GCP and Azure marketplaces at the same $10 per million input tokens and $50 per million output tokens that Fable 5 charged.
Claude Mythos 5.1 launched the same day as a separate, more capable model, but Anthropic did not put it on general release. It sits behind two vetting programs, and until an organization clears one of them, Mythos 5.1 is not available at any price.
Mythos 5.1's gate: two verification programs, one border
Claude Mythos 5.1 access runs through Anthropic's Cyber Verification Program and Life Sciences Verification Program, and both programs currently accept applications from US-based organizations only.
The programs themselves are scoped to the United States: an organization applies, Anthropic vets it against one of two specific use-case tracks, and only a US-based applicant currently has anything to file. An EU cyber-defense team or a European life-sciences research group evaluating Mythos 5.1 today has no application path at all.
The benchmark jump, in three numbers
Fable 5.1 scored 52.6 percent on Terminal-Bench-Science, more than double the 24.7 percent Fable 5 posted on the same benchmark, and it reached 55.8 percent on the broader Terminal-Bench 4.0.
| Fable 5.1 vs Fable 5 - Terminal-Bench-Science | 52.6% vs 24.7% |
|---|---|
| Fable 5.1 - Terminal-Bench 4.0 | 55.8% |
| Fable cache-read price, before to after | $1.00 to $0.25 per million tokens (-75%) |
| Fable 5.1 pricing (input/output) | $10 / $50 per million tokens, unchanged |
| Fable 5.1 access | General availability - API, AWS, GCP, Azure |
| Mythos 5.1 access | Vetted orgs only - Cyber + Life Sciences Verification Programs - US only |
Anthropic did not publish comparable Terminal-Bench-Science or Terminal-Bench 4.0 scores for Mythos 5.1 alongside the Fable numbers, which leaves EU teams unable to size the capability gap between the model they can buy and the one they cannot, beyond knowing Mythos sits above Fable in Anthropic's own lineup.
What the cache-read cut is worth in practice
Anthropic cut prompt cache-read pricing for Fable from $1.00 to $0.25 per million tokens, a 75 percent reduction that lands directly on any workload built around repeated context.
A retrieval-augmented generation pipeline or an agent that re-reads the same system prompt, document set or tool schema across hundreds of calls a day pays for that context on every cache read, not just once. At the old rate, a workload re-reading 50 million cached tokens a day cost $50 in cache reads alone, roughly EUR 46; at the new rate it costs $12.50, roughly EUR 11.50, a saving worth totaling before any EU team writes off Fable as the compromise tier.
The roadmap decision this creates for EU buyers
An EU procurement team building a frontier-tier roadmap for cyber-defense or life-sciences workloads now has to choose between two paths that were not both open before September 1.
One path is to plan the roadmap around Mythos 5.1 anyway, treating today's access gap as temporary and betting that Anthropic extends the Cyber Verification Program or the Life Sciences Verification Program to EU organizations before the capability is actually needed in production. The other is to route the budget to Fable 5.1's now cheaper, generally available tier, build the workload there, and accept a capability lag against US-based competitors who can already apply for Mythos 5.1 today.
Neither path is free. The first risks a roadmap built on a program that may never open to European applicants on the timeline the business case assumes. The second is a known, bookable decision: Fable 5.1's benchmark gains and lower cache-read cost are real today, not contingent on a vetting process outside the buyer's control.
A dated point in a widening access gap
Frontier labs have restricted their most capable models to a single jurisdiction before; the Mythos 5.1 launch gives that pattern a concrete date, September 1, 2026, and a named mechanism, two verification programs scoped to the United States.
For an EU enterprise, the practical lesson is narrower than the headline: frontier-tier access can now split by jurisdiction on a named legal or regulatory basis rather than by infrastructure readiness, and a procurement roadmap that assumes uniform global access to whichever model tests best is no longer a safe assumption to build on.
Servola Journal
We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.
Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.
If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.
Read next: Cache Reads Drop 75%, Data Retention Rule Reverses | Carnegie Mellon Wired a Lab Robot to Claude in 8 Hours
