“Artificial Analysis found that due to high token generation, running its entire benchmark suite with Claude Sonnet 5 was more expensive than running it with Fable-5.”