BigQuery on-demand vs editions: when paying for slots is cheaper

BigQuery on-demand charges for the data your queries scan: $6.25 per TiB in the US, with the first TiB each month free. BigQuery editions charge for compute, in slot-hours: $0.04 on Standard, $0.06 on Enterprise and $0.10 on Enterprise Plus, whatever the queries scan. So the cheaper model depends on 1 ratio: how many slot-hours your queries use for each TiB they scan. Below 156 slot-hours per TiB, Standard is cheaper; below 104, Enterprise is. You can measure the ratio from your own query history in a few minutes.

Where the break-even comes from

One TiB scanned on demand costs $6.25. The same $6.25 buys:

A query that scans a lot but does little with it, such as a filter over a wide table, uses few slot-hours per TiB, so slots win. A query that scans little but joins, sorts and aggregates heavily uses many, so on-demand wins. Most teams have both kinds, which is why the answer has to come from their own history.

Move the 2 numbers for your month and see which side wins:

Try itBigQuery: on demand or slots for your month?

Scanned a month

30TiB

Slot-hours a month

2,000

• On demand$181
• StandardCheapest$80
• Enterprise$120
• Enterprise Plus$200

Slot-hours per TiB scanned

15610462.5

← slots winYou: 67on demand wins →

Dashed lines: below 156 Standard beats on demand, below 104 Enterprise, below 62.5 Enterprise Plus.

Measure your own ratio

Every query's slot use and billed bytes are in the INFORMATION_SCHEMA.JOBS view, kept for 180 days:

SELECT
  SUM(total_bytes_billed) / POW(1024, 4)          AS tib_billed,
  SUM(total_slot_ms) / (1000 * 60 * 60)           AS slot_hours,
  SAFE_DIVIDE(SUM(total_slot_ms) / 3600000,
              SUM(total_bytes_billed) / POW(1024, 4)) AS slot_hours_per_tib
FROM `region-us`.INFORMATION_SCHEMA.JOBS
WHERE job_type = 'QUERY'
  AND creation_time >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 30 DAY);

Run it per project, and again grouped by user or service account: a single dashboard that refreshes every 5 minutes can decide the answer on its own.

2 teams, opposite answers

The reporting team. 30 TiB scanned in a month, 2,000 slot-hours used: about 67 slot-hours per TiB.

• On demand: (30 − 1 free) × $6.25 = $181.25
• Standard: 2,000 × $0.04 = $80, Enterprise: $120

Slots are cheaper, even before commitments.

The data science team. 200 TiB scanned, 40,000 slot-hours of heavy joins and window functions: 200 slot-hours per TiB.

• On demand: 199 × $6.25 = $1,243.75
• Standard: 40,000 × $0.04 = $1,600, Enterprise: $2,400

On-demand is cheaper, and moving to slots would raise the bill by a third or more.

The margin you need

The measured slot-hours are what queries used. An autoscaling reservation bills for what it holds:

• It grows in steps of 50 slots and keeps new capacity for at least 60 seconds, so a 5-second query can pay for 50 slots for a minute.
• Billing is per second with a 1-minute minimum.
• Baseline slots are billed all the time, busy or idle.

So treat the measured ratio as a floor and add a margin. If slots only win by 10% on paper, stay on demand. If they win by half, as for the reporting team, switch, set the autoscaler's maximum, and compare the first full month's bill with your estimate. Idle baseline slots can be shared with other reservations in the same admin project, which helps a company with several teams more than a single one.

Other things that change the answer

• Commitments: 1 or 3-year commitments on Enterprise and Enterprise Plus lower the slot-hour price, which moves the break-even up. Only commit to a baseline you use every hour.
• Scanning less helps both: partitioning and clustering cut bytes on demand and slot time under editions alike. Do that first; the partitioningData Partitioning and ClusteringTwo ways to lay out a large table so a query reads only the files it needs: partitioning splits it by a column such as the date, clustering keeps similar values in the same files. concept page explains why.
• The free tier: the first TiB each month is free only on demand, which matters to small projects and not at all to large ones.
• Concurrency: on-demand projects share up to 2,000 slots. If heavy queries queue at busy hours, a reservation also buys predictability, not only price.

Our BigQuery cost calculator prices both models in every region, with storage and streaming, and Snowflake vs BigQuery sets the same workload against a Snowflake warehouseData WarehouseA database built for analysis: cleaned, structured data from many systems, kept in one place and shaped for fast reports and dashboards..