In a post on X, Boris Cherny outlined several additional considerations around Anthropic's Claude Fable 5.1 and Claude Mythos 5.1, including pricing, safeguards, benchmark results, and usage settings.
Cherny claims that its latest biology safeguards intervene on benign requests "85% less often" than the safeguards shipped with Fable 5. Claude Code users are expected to see "roughly 60% fewer cyber-safety interventions per session." Those figures are company-provided comparisons rather than independently verified measurements.
An included Terminal-Bench 4.0 chart compares Mythos 5.1, Fable 5.1, and Mythos 5 across five effort settings. Mythos 5.1 scores about 42% at low effort, 47% at medium, 57% at high, 60% at xhigh, and 61% at max effort. Fable 5.1 scores roughly 40%, 44%, 49%, 51%, and 56%, respectively.
The chart also lists mean costs per task on a logarithmic scale. Mythos 5.1 ranges from about $6 at low effort to $18 at max effort, while Fable 5.1 ranges from roughly $6.50 to $19.50. The earlier Mythos 5 ranges from about $12.50 to $23.50 across the same settings.
Cherny also described Fable 5.1 as having improved writing and tone. He noted that Anthropic is still working to reduce what he called "Claude-speak," suggesting the stylistic changes remain in progress.
In a follow-up, Cherny mentioned using the xhigh setting, while noting that many people prefer medium because it uses fewer tokens and has lower latency. He also attributed at least one reported issue to a configuration-cache freshness problem and recommended restarting Claude.
Source: Boris Cherny on X
