Anthropic released Claude Fable 5.2 on November 9 and reported higher scores than Opus 5.5 on several internal coding benchmarks while recommending Opus as the daily driver. The New Stack author then ran both models on three coding and agent tasks at default API settings without additional tuning. Fable used high effort by default while Opus used medium effort.
Across nine total runs the models produced identical perfect scores. Opus 5.5 finished every version faster and used fewer reasoning tokens. Total estimated API cost for all runs was $2.29 for Opus versus $8.98 for Fable. Total runtime was 701 seconds for Opus versus 1,579 seconds for Fable.
The tests covered a cron scheduler with time-zone rules, an invoicing library bug hunt, and an incident log investigation. Each task started with a baseline version and was made progressively harder. The author notes that the small scale prevents direct comparison to Anthropic’s larger benchmark numbers and that different effort settings may explain part of the observed gap.
Anthropic’s reported benchmark scores for Fable were measured at high effort while the independent runs used each model’s default configuration. The New Stack results therefore reflect what a developer receives without manual prompt or effort tuning.