Serverless DBU consumption by SKU
This article explains the SKUs and DBU multipliers used to bill for various Databricks serverless offerings.
:::
What is a DBU multiplier?
When using certain features, a multiplier is applied to the underlying DBUs consumed. For instance, Lakehouse Monitoring has a 2X multiplier. If the associated background job uses 5 DBUs, you are billed for 10 DBUs after applying the multiplier. The DBUs shown on your bill and in system tables reflect the final amount after this multiplier is applied. See What is a DBU for the definition of a DBU.
Automated Serverless SKU
The following capabilities are billed against the Automated Serverless SKU.
Feature | DBU multiplier |
|---|---|
Serverless Jobs | 1X |
Serverless Lakeflow Declarative Pipelines | 1X |
Serverless Lakeflow Declarative Pipelines with Advanced Pipeline Features | 1.5X |
Predictive Optimization | 1X |
Lakehouse Monitoring | 2X |
Fine-Grained Access Control (Preview) | 1X |
Online tables synchronization (Preview) | 1X |
Online tables synchronization with Advanced Pipeline Features (Preview) | 1.5X |
Online tables Capacity Unit (Preview) | 2X |
Materialized Views and Streaming Tables in Databricks SQL | 1X |
Materialized Views and Streaming Tables in Databricks SQL with Advanced Pipeline Features | 1.5X |
Interactive Serverless SKU
The following capabilities are billed against the Interactive Serverless SKU.
Product / Feature | DBU Multiplier |
|---|---|
Serverless Notebook | 1X |
Databricks App capacity hour | 0.5X |
SQL Serverless SKU
The following capabilities are billed against the SQL Serverless SKU.
Product / Feature | DBU Multiplier |
|---|---|
Warehouse Size | DBU/hour |
2X-Small | 4 |
X-Small | 6 |
Small | 12 |
Medium | 24 |
Large | 40 |
X-Large | 80 |
2X-Large | 144 |
3X-Large | 272 |
4X-Large | 528 |
Model Serving SKU
The following capabilities are billed against the Serverless Real-Time Inference SKU.
AI Gateway
Product / Feature | DBU Multiplier |
|---|---|
AI Guardrails | 21.429 DBUs / M tokens |
Inference Tables for CPU, GPU endpoints | 7.143 DBUs / 1 GB of payload |
Usage Tracking for CPU, GPU endpoints | 1.429 DBUs / 1 GB of payload |
CPU Model Serving
1 concurrent request/hr = 1 DBU/hr
GPU Model Serving
Instance Size | GPU configuration | DBUs / hour |
|---|---|---|
Small | T4 or equivalent | 10.48 |
Medium | A10G x 1GPU or equivalent | 20.00 |
Medium 4X | A10G x 4GPU or equivalent | 112.00 |
Medium 8x | A10G x 8GPU or equivalent | 290.80 |
XLarge | A100 40GB x 8GPU or equivalent | 538.40 |
XLarge | A100 80GB x 8GPU or equivalent | 628.00 |
Foundation Model Serving
Model | Pay-Per-Token | Provisioned Throughput | |
|---|---|---|---|
DBU / 1M INPUT tokens | DBU / 1M OUTPUT tokens | DBU per hour | |
Current Models | |||
Llama 4 Maverick | 7.143 | 21.429 | 85.715 |
Llama 3.1 405B | 71.429 | 214.286 | 700.000 |
Llama 3.1 70B | |||
Llama 3.1 8B | n/a | n/a | 106.000 |
Llama 3.2 3B | n/a | n/a | 92.857 |
Llama 3.2 1B | n/a | n/a | 85.714 |
DBRX | 10.714 | 32.143 | 171.429 |
Mixtral 8x7B | 7.143 | 14.286 | 157.143 |
GTE | 1.857 | n/a | 20.000 |
BGE Large | 1.429 | n/a | 24.000 |
Legacy Models | |||
Llama 3 70B | n/a | n/a | 212.143 |
Llama 3 8B | n/a | n/a | 106.000 |
Llama 2 70B | 7.143 | 21.429 | 157.143 |
Llama 2 13B | n/a | n/a | 78.571 |
MPT 30B | n/a | n/a | 112.000 |
MPT 7B | n/a | n/a | 20.000 |
GTE | 1.857 | 1.857 | n/a |
BGE Large | 1.429 | 1.429 | 10.480 |
Anthropic Model Serving
Model | Pay-Per-Token | Provisioned Throughput | |
|---|---|---|---|
DBU / 1M INPUT tokens | DBU / 1M OUTPUT tokens | DBU per hour | |
Claude Opus 4 | 214.286 | 1,071.429 | |
Claude Sonnet 4 | 42.857 | 214.286 | |
Claude Sonnet 3.7 | 42.857 | 214.286 |
Note: A promotional price reduction of 16.7% will be applied to the DBU rates shown in the table until September 31, 2025
Shutterstock Image AI
1 image = 0.857 DBUs
Vector Search
DBU/hour for 1 unit | Vector Capacity per Unit | |
|---|---|---|
Standard | 4.0 | 2 million |
Storage optimized | 18.29 | 64 million |
Agent Evaluation
1 judge request = 1 DBU
Model Training
The following capabilities are billed against the Model Training SKU.
Model Training - Fine Tuning
Model | Training word count | Approximate DBUs |
|---|---|---|
Current Models | ||
Llama 3.1 405B | 10,000,000 | 1,150 |
500,000,000 | 57,150 | |
Llama 3.1 70B | 10,000,000 | 375 |
500,000,000 | 17,600 | |
Llama 3.1 8B | 10,000,000 | 150 |
500,000,000 | 6,600 | |
Llama 3.2 3B | 10,000,000 | 75 |
500,000,000 | 3,300 | |
Llama 3.2 1B | 10,000,000 | 25 |
500,000,000 | 1,100 | |
DBRX | 10,000,000 | 300 |
500,000,000 | 14,300 | |
Mixtral 8x7B | 10,000,000 | 150 |
500,000,000 | 6,600 | |
Mistral 7B | 10,000,000 | 50 |
500,000,000 | 1,325 | |
Legacy Models (to be deprecated on Dec 13, 2024) | ||
Llama 3 70B | 10,000,000 | 375 |
500,000,000 | 17,600 | |
Llama 3 8B | 10,000,000 | 150 |
500,000,000 | 6,600 | |
Llama 2 70B | 10,000,000 | 275 |
500,000,000 | 13,200 | |
Llama 2 13B | 10,000,000 | 50 |
500,000,000 | 2,475 | |
Llama 2 7B | 10,000,000 | 25 |
500,000,000 | 1,175 | |
Codellama 34B | 10,000,000 | 100 |
500,000,000 | 4,950 | |
Codellama 13B | 10,000,000 | 75 |
500,000,000 | 2,650 | |
Codellama 7B | 10,000,000 | 50 |
500,000,000 | 1,325 |
Databricks Storage
The following capabilities are billed against the Databricks Storage SKU
Product / Feature | DSU Multiplier |
|---|---|
Vector Search | 10X |
Online Tables Storage (preview) | 15X |