Claude Fable 5.1

Claude Opus 5's capabilities are close to the level of the previous-generation flagship Opus, but with significantly reduced inference costs.

The core difference between Fable 5.1 and Mythos 5.1 lies in the depth of safety measures, not in the underlying capability level.

In an AI travel website landing page frontend design test, the reviewer judged GPT-6 Astra to outperform Fable 5.1.

The name "Claude Fable 5.1" may be the community's designation for a certain Claude variant or internal version.

In actual testing, GPT-6 Astra generated a first-person shooter game prototype featuring container bunkers, fully automatic rifles, and recoil and ballistic determination within half an hour, while the control group Claude Fable 5.1 took close to 5 hours.

In the Terminal Bench 4.0 highest-accuracy test, Fable 5.1 achieved an accuracy of 55.8% at a cost of $19.50, while Astra achieved an accuracy of 56.7% at a cost of $10.35.

In its product description, Anthropic states that Fable 5.1's research capabilities offer an early glimpse into how AI models can contribute to scientific progress.

Anthropic's status page recorded an incident involving Claude Mythos 5.1 and Claude Fable 5.1, in which both models experienced intermittent increases in error rates during a similar time period.

Incident timeline: first marked as Investigating at 09:48 UTC, updated at 09:52 UTC, moved to Monitoring at 10:50 UTC, and marked as Resolved at 11:14 UTC.

Meta's Alexandr Wang has said API pricing would be 'significantly lower' than competitors, which holds for Claude Fable 5.1's $10/$50 per million tokens pricing but does not hold for Claude Opus 5.5.

Fable 5.1 and Opus 5.5 lead on surprise correlation.

Claude Fable 5.1 launched with a cache price of 2.5%, Claude Opus 5.5 and GPT-6.1 Sol both at 5%, and by October 7 Anthropic also halved the Sonnet 5.5 cache price to 5%.

Opus 5.5 is not Anthropic's flagship model; the true flagship is Claude Fable 5.1.

In Every team's tests, 12 passages had errors planted in them; Jev caught 6 out of 7 in 0.35 seconds, Claude Fable 5.1 caught all 7 but took 8.83 seconds, and Jev was approximately 580x cheaper.

In the experimental setup, the four Claude agents had a maximum of 40 rounds of conversation to identify the infiltrating Kimi, while Kimi's task was to avoid being detected.

All players chose Samwise (a character from The Lord of the Rings) on the coffee companion prompt, indicating that models are difficult to distinguish on questions with high consensus.

On the color prompt, all four Claude agents chose teal, while Kimi chose blue.

In the third set of tests, all four Claude agents chose 37 on the random number prompt, while Kimi chose 73.

Kimi's failure was due to a tendency to ask open-ended reflective questions, combined with a failure to consistently use the commit-reveal sealed mechanism, which allowed Claude to answer later and blend into the group.

The Databricks model library includes Claude Opus 5.5, Claude Fable 5.1, and more than 60 open-source and frontier models.