Note: For image generation models such as SDXL, the pricing is based on the number of inference steps, which refers to the denoising iterations involved in the image creation process. All the FLUX models share the same pricing structure.
The pricing for all FLUX models is based on a standard number of processing steps. Additionally, users should be aware that more steps can enhance the quality and detail of the generated images, making it important to balance cost with desired output quality.

Note:For speech-to-text models, we bill based on the duration of audio input, charging per second. This pricing structure allows users to efficiently manage costs based on the length of the audio they wish to transcribe.

Note: The pricing for embedding models is determined by the quantity of input tokens that the model processes. This means that the cost will vary depending on the length and complexity of the text being analyzed. It means more tokens lead to higher costs.

Note: The charges are based on the total number of tokens in your fine-tuning dataset, calculated as the dataset size multiplied by the number of epochs. The bill is only for the fine-tuning process itself with no additional fees for deploying fine-tuned models, and the inference costs remain the same as those for the base model. Users can deploy and manage multiple fine-tuned models without incurring extra costs.

Note: On-demand deployments are charged based on GPU usage measured in GPU-seconds. The pricing increases proportionally when utilizing multiple GPUs. Additionally, users can optimize their costs by selecting the appropriate model based on their specific workload requirements and budget constraints.