Pricing

Question 1

Pricing

Charge model for text prompting and output?

Answer 1

Charged per Token in and Token out

Question 2

Pricing

Charge model for embedding models?

Answer 2

Charged for every input token processed

Question 3

Pricing

Charge model for image models?

Answer 3

Charge for each image generated

Question 4

Pricing

How is Batch inference price comapred to non-batch?

Answer 4

Batch can be up to 50% cheaper

Question 5

Pricing

How does Provisioned work?

Answer 5

Purchase model units for a certain time (1 month, 6 months, …)

Question 6

Pricing

What do you get with Provisioned Throughput?

Answer 6

Guaranteed throughput (max tokens in/out per minute)

Question 7

Pricing

What types of models support Provisioned Throughput?

Answer 7

Base, Fine-tuned, and custom models

Question 8

Pricing

Cheapest of the Model Improvement techniques?

Answer 8

Prompt engineering – no training, nothing special

Question 9

Pricing

Next-cheapest of the Model Improvement techniques?

Answer 9

RAG: model doesn’t have to know everything, stored externally

Question 10

Pricing

Next cheapest of the model improvement techniques after RAG?

Answer 10

Instruction-based fine-tuning

Question 11

Pricing

Most expensive of the Model Improvement techniques?

Answer 11

Domain adaptation fine-tuning

Question 12

Pricing

How can Provisioned Throughput save costs?

Answer 12

Can’t, it’s reserved capacity

Question 13

Pricing

How can Temperature save costs?

Answer 13

Can’t

Question 14

Pricing

How can Top-K / Top-P save costs?

Answer 14

Can’t

Question 15

Pricing

Main cost driver for LLMs?

Answer 15

Number of input and output tokens