Frontier AI models beat human analysts at earnings prediction

Samaya built a finance-specific prediction harness and tested frontier AI models against analyst consensus on 456 Q2 earnings releases. The latest models outperformed both raw and bias-corrected consensus in predicting revenue surprises.

• Evaluated on 456 quarterly earnings releases across four headline metrics
• Prediction error falls as model capability increases
• Models beat both raw and bias-corrected analyst consensus
• Predictions made five trading sessions before each earnings release