Contract Law: How Tokens are Driving New Contract Considerations

Tokens aren’t just for the arcade these days. AI systems utilize “tokens” for pricing and usage. Though the concept is similar to its namesake, AI users purchase tokens, and in return, they are allotted a certain amount of usage. Tokens represent a measure of compute. The more complex the task or the AI model used, the higher the token cost. More frequently,  we’re seeing AI companies shift enterprise models from unlimited usage for a flat fee to token-based pricing. This raises considerations for contract drafters who must discern how token licensing provisions will impact their clients and negotiate the best deal. A recent Venable memo outlines some of these considerations and how they impact enterprise AI use:

“Token licensing also raises questions that traditional software agreements rarely address. How are tokens measured and reported? Can the provider change token consumption rates for particular models? What happens when a provider introduces a new model that consumes tokens differently from its predecessor? Are purchased tokens refundable if the service is terminated? These questions often have a direct financial impact and should be evaluated alongside more familiar provisions relating to fees, service levels, and termination rights.”

Tokens can make AI pricing opaque and costs difficult to predict. There is no standard number of tokens used per AI prompt. Token usage boils down to specifics. These include the length of the prompt, the parameters of the response, and the AI model used. Keeping tabs on token usage is critical to controlling costs.