article --summary
The summary
The author ran both models through the same three coding tasks: fixing planted bugs in a billing codebase, reviewing an invoicing package for defects, and writing validation helpers from scratch. Both models tied on bug fixing and code writing, while Fable 5.2 led only on the code review, finding one more defect than Astra and raising no false alarms.
The main lesson is about cost rather than scores. Fable 5.2 used far more tokens, so its lower standard rates did not keep its total bill down, and it cost roughly 2.4 times as much as Astra on the two tied tasks and took nearly twice as long. The author suggests Astra for routine work and Fable 5.2 for careful reviews where a missed rule could cause real harm, while noting that each model was run only once.
Editorial note: This is a summary of an article first published by The New Stack. The full piece is on their site.