MaxCompute:Compute pricing (pay-as-you-go)

How billing works

Fee = Input data scanned (GB) × SQL complexity × USD 0.0438

Input data scanned is the compressed data size that the SQL job actually reads—not the full table size. Two optimizations reduce this:

• Partition filtering: If your query contains WHERE ds > 20130101 and ds is a partition key, only the data in matching partitions is billed.

• Column pruning: SELECT f1, f2, f3 FROM t1 bills only columns f1, f2, and f3—not the entire table.

SQL complexity is based on the number of SQL keywords in the statement:

The keyword count formula is:

Keywords = JOINs + GROUP BYs + ORDER BYs + DISTINCTs + window functions
           + MAX(INSERTs | UPDATEs | DELETEs − 1, 1)

To estimate cost before running a query, use the COST SQL command:

COST SQL <your SQL statement>;

Unit price

Calculation examples

Example 1: A query with DISTINCT, GROUP BY, and ORDER BY

odps@ $odps_project >COST SQL SELECT DISTINCT total1 FROM
(SELECT id1, COUNT(f1) AS total1 FROM in1 GROUP BY id1) tmp1
ORDER BY total1 DESC LIMIT 100;
Input: 1825361100.8 Bytes
Complexity: 1.5

This query has 4 keywords (1 DISTINCT + 1 GROUP BY + 1 ORDER BY + MAX(0−1, 1) = 1), giving a complexity of 1.5. With ~1.7 GB of input data:

Cost = 1.7 GB × 1.5 × USD 0.0438 = USD 0.11

Example 2: Effect of column pruning on cost

Suppose a table has 20 columns and is 40 GB compressed. A full-table scan bills the entire 40 GB. If your query reads only 3 of those 20 columns (~6 GB of compressed data), the cost drops proportionally:

Selecting only the columns you need can reduce SQL costs significantly.

How billing works

Fee = Input data volume (GB) × USD 0.0044

Key differences from standard SQL billing:

• SQL complexity is always 1 (not calculated).

• When a query joins both internal and external tables, each part is billed at its respective rate.

• You cannot estimate fees for external table queries before running them.

Unit price

How billing works

Daily fee = Total billable hours × USD 0.0690

Billable hours for a single job:

Billable hours = Job runtime (hours) × Number of CPU cores used

For example, a job using 100 cores for 30 minutes: 100 × 0.5 = 50 billable hours.

Unit price

How billing works

Daily fee = Total billable hours × USD 0.1041

Billable hours take the larger of the CPU-based and memory-based calculations:

Billable hours = MAX(CPU cores × runtime hours, CEILING(Memory GB × runtime hours / 4))

This ensures you are billed for whichever resource—CPU or memory—your job uses more of.

Examples:

In the second example, memory drives the cost despite the same number of CPU cores.

Unit price

How billing works

Mars jobs use the same billing formula as Spark jobs:

Daily fee = Total billable hours × USD 0.1041

Billable hours = MAX(CPU cores × runtime hours, CEILING(Memory GB × runtime hours / 4))

Examples:

Unit price

How billing works

Fee = Input data scanned (GB) × SQL complexity × USD 0.0438

MCQA uses columnar storage with data compression. Charges are based on compressed data size.

If you have purchased subscription resources and allocated a portion to MCQA, jobs running within that quota are not billed additionally. If you use only pay-as-you-go, MCQA jobs are billed the same way as standard SQL jobs.

Partition filtering reduces the amount of scanned data and lowers costs, the same way it works for standard SQL jobs.

Unit price

How billing works

Daily fee = Total billable hours (CU-hours) × USD 0.0541

Unit price

Standard MaxFrame jobs (using the pay-as-you-go Standard Edition computing quota):