GPT 6 Astra vs Fable 5.1: Too Close to Call, But Dual-Model Cross-Critique Wins

peterwildeford · x · 2026-09-07

Peter Wildeford, after two days of heavy use, finds GPT 6 Astra not meaningfully better than Fable 5.1 for his work (policy analysis, memo writing, conceptually tricky software like dashboards and forecasting models)—but using both side by side is highly additive. Which model wins is fairly random per task. For the hardest conceptual tasks, having each model do the project and then critique each other produced far better output than either alone. A reasonable person could call either model 'best,' he concludes.

Original post →

More from Models

Models channel →