Frontier AI models with Samaya’s harness reportedly beat expert consensus on earnings predictions

A post linking to Samaya’s research says Opus 5.5, Fable 5.1 and Astra, using Samaya’s harness, meaningfully outperformed bias-corrected expert consensus on earnings predictions, including revenue and gross margin. The researchers say they designed a point-in-time gate to prevent information leakage and gave the models access to the same information as experts. They claim this is the first time frontier AI models have outperformed human experts on financial predictions.