SaaS firms weigh hybrid pricing to recover rising LLM token costs

Many SaaS companies are struggling to recover the variable costs of large language model features, which are billed per token and can vary widely by model and usage. A worked example using flagship OpenAI rates shows a heavy user costing roughly $73 per month against a $40 seat price, illustrating why flat pricing can lose money. Vendors increasingly favor hybrid models that combine a subscription with an AI allowance, metered overage, or usage caps. The recommended first step is routing all model calls through a gateway that tags requests by tenant and feature and logs provider-reported token counts at the rates in effect at the time. Teams are advised to gather at least one full billing cycle of this per-tenant cost data before changing prices.