Hidden costs in AI deployment: Why Claude models may be 20-30% more expensive than GPT in enterprise settings
1 min read
Summary
Tokenisation varies according to the model and the model family, and this has implications for the number of tokens generated for any given text, and the costs incurred as a result.
OpenAI’s ChatGPT and Anthropic’s Claude 3.5 Sonnet offer competitive pricing for output tokens, but Claude offers a lower cost for input tokens.
However, experiments reveal that, despite the lower input token rates of the Anthropic model, the total cost of running experiments for both models is much higher for Claude.
This is because the Anthropic tokeniser breaks down the same input into more tokens compared to OpenAI’s tokeniser, meaning that Anthropic models produce significantly more tokens than OpenAI counterparts, pushing up processing costs.
The extent of the problem varies according to the domain, with technical documentation and code yielding the most significant differences in costs.
This highlights the need to consider the nature of the text when companies are choosing between OpenAI and Anthropic models for natural language processing tasks.