Anthropic, a company known for its generative AI model called Claude, offers four distinct models: Haiku, Sonnet, Opus, and Fable. Each is tailored for different tasks, ranging from quick responses to complex reasoning. Haiku is the most affordable and lightweight, ideal for simple tasks like summaries or quick answers. Sonnet is more powerful, suited for coding, writing, and multi-step tasks. Opus is even heavier, designed for deep research and complex reasoning. Fable, the most expensive, is meant for critical and long-term projects requiring substantial computational power. To understand the cost differences, Next conducted extensive testing on nearly 200 combinations of models and prompts. The tests used the Anthropic API, which bills users based on the number of input and output tokens. This is different from Anthropic’s Pro and Max packages, which have vague usage limits that can be confusing for users. The API’s clear pricing model allows for precise cost analysis, which is what Next aimed to provide. The tests involved three distinct prompts: a riddle, a travel journal request, and a Chrome/Firefox extension design. Each prompt was run across 11 different models, including various versions of Haiku, Sonnet, Opus, and Fable. The results showed a wide range in costs. Haiku was the cheapest, averaging 1.3 cents per request, while Fable was the most expensive, averaging 60 cents. Sonnet and Opus fell in between, with Sonnet averaging 15 cents and Opus 20 cents. The cost differences are significant: Sonnet is about 11 times more expensive than Haiku, Opus is 40% more than Sonnet, and Fable is three times more than Opus. The latest versions of the models showed even greater cost differences. Haiku 4.5 remained the cheapest at 1.3 cents, while Sonnet 5 averaged 19.3 cents, Opus 5 averaged 27.8 cents, and Fable 5.1 reached 92.1 cents. This means the price gap between Haiku and Fable 5.1 is nearly 100 times. These results highlight the importance of choosing the right model based on both the task and budget. Next will continue to explore these findings in more detail in upcoming articles.