Two days after we launched @Signal_65 PINNACLE, Anthropic shipped the model that shows why we built it.
Claude Fable 5.1 fails about half as many agent workflows as anything we had measured, and it makes up an answer one time in 140 instead of one in 13. That is the number that