Generative AI Service Pricing

Foundational models can be consumed on demand, where you pay per character based on the length of the prompt and the response from the model (except for the embedding models, where the response from the model isn’t accounted for). In the table below, a transaction = a character and 10,000 transactions = 10,000 characters.

Additionally, you can host private replicas of foundational models and create fine-tuned models on dedicated AI clusters. Dedicated AI clusters come in two types: hosting and fine-tuning. You create a hosting cluster by assigning AI units to it based on the model you want to host and the expected call volume to the model. Fine-tuning clusters require two AI units of the specific model you want to fine-tune. Once you create a fine-tuned model in a fine-tuning cluster, you can host it on your hosting cluster.

Dedicated AI clusters require a minimum commitment of 744 unit-hours (per cluster) for hosting models. Fine-tuning clusters require a minimum of 1 unit-hour.