article --summary
The summary
The author asks why Anthropic would recommend its cheaper Opus model over its newer Fable 5.2 release, and whether the cheaper option is good enough. Anthropic's own benchmarks show Fable ahead, but the author ran a smaller set of three hard tasks against both models at default API settings.
Across three versions each of a cron scheduler, an invoicing bug hunt, and a payments outage investigation, both models scored perfectly. Fable used considerably more reasoning and cost roughly four times as much overall, while taking more than twice as long. The author notes that the default effort settings differ between the two models, so the comparison does not fully isolate the model itself.
The conclusion is to begin with Opus 5.5 and move to Fable only when Opus fails to produce satisfactory results. The author stresses that these tests were short and do not disprove Anthropic's broader claims.
Editorial note: This is a summary of an article first published by The New Stack. The full piece is on their site.