I just used Claude Fable and wasted over $20 solely on its thinking process— I wouldn't recommend it right now.
The model's API pricing is incredibly high. When I asked it to debug a performance issue in my project, it "thought" for over 10 minutes and racked up a $20+ bill. After all that, it flagged my prompt as a "cybersecurity violation" and redirected me to Opus.
By that point, it hadn't generated a single meaningful output, and I highly doubt Fable's thinking process would have actually helped with the task anyway.
That’s really dumb.