How much does a token cost? (www.erikjs.com)

A recent analysis has quantified the resource costs associated with using the GPT-4o model for token processing, specifically for converting the term "strawberry" into tokens. The process requires approximately 3 tokens and consumes minimal resources: around 960 KB of GPU memory, 0.002 Wh of energy, and 0.01 mL of water. Financially, the input cost for processing is estimated at just $0.00001 per token, making it a highly efficient option for AI applications. This insight holds significant implications for the AI/ML community as it highlights the low environmental and operational costs associated with large language model usage, crucial for sustainability discussions. With the potential for processing around one million tokens daily, the efficiency becomes even more pronounced, particularly in terms of energy and resource consumption. The data supports ongoing efforts to optimize AI models not only for performance but also for their ecological footprint, paving the way for more sustainable AI practices in the future.