Gemini in BigQuery Pricing for Conversational Analytics and Data Agents
The coding help is free. The agents that answer business questions are free only until 31 December 2026, and Data Insights starts billing on 27 October. Here is what each meter costs and how to budget it.
Checked 15 October 2026 on cloud.google.com and Google Cloud docs · every rate quoted
Ask the demo shop a question, or type your own:
Reading the schema, writing SQL… Writing SQL… Running (read‑only)… SQL AgentSQL wrote
▋
A real MySQL shop: customers, products, orders and signups over the last year. AgentSQL writes the SQL, runs it read-only, and answers.
Direct answer
Gemini in BigQuery has two price tiers. SQL code assist, Python code assist, data canvas and data preparation are included at no extra cost on every BigQuery edition and on demand. The features that answer questions for business users are metered: the Conversational Analytics Agent, the Data Science and Data Engineering agents and Data Insights all bill $3 per million input data tokens and $20 per million output data tokens. Data Insights billing starts on 27 October 2026 and the agents' free trial ends on 31 December 2026. The queries they run are billed as normal BigQuery compute, $6.25 per TiB on demand, from the first day.
›_ gemini in bigquery pricing, feature by feature
Eight things Gemini does in BigQuery and what each one costs.
Read from Google's Gemini pricing overview, the Data Cloud agents pricing page and the BigQuery pricing page on 15 October 2026.
| Feature | Today | After the free period | Google source |
|---|---|---|---|
| SQL code assist (generate, explain, complete) | Included | Included on every edition and on demand | Gemini pricing overview |
| Python code assist | Included | Included on every edition and on demand | Gemini pricing overview |
| Data canvas | Included | Included on every edition and on demand | Gemini pricing overview |
| Data preparation | Included | Included on every edition and on demand | Gemini pricing overview |
| Data Insights (table insights and generated queries) | Free until 27 Oct 2026 | $3 per 1M input and $20 per 1M output data tokens | Data Cloud agents pricing, Dataplex billing FAQ |
| Conversational Analytics Agent | Free trial to 31 Dec 2026 | $3 per 1M input and $20 per 1M output data tokens | Data Cloud agents pricing |
| Data Science Agent and Data Engineering Agent | Free trial to 31 Dec 2026 | $3 per 1M input and $20 per 1M output data tokens | Data Cloud agents pricing |
| The queries any of these run | Billed now | $6.25 per TiB on demand after 1 free TiB a month, or slot-hours on an edition | BigQuery pricing page |
The split is deliberate. Google gives away the features that help someone who already writes SQL work faster inside BigQuery Studio, and meters the features that let someone who does not write SQL get an answer. If your plan is to give sales, ops or finance people a way to ask BigQuery questions in plain English, you are on the metered side of that line. The full compute side of the bill, editions, slot-hours and storage, is on our BigQuery pricing page.
›_ when the free period ends
Two dates decide your first Gemini bill.
Each date is quoted from a Google page, with the exact wording below.
27 Oct 2026
Data Insights bills
"Data Insights pricing starts on October 27, 2026." Twelve days after we read it.
31 Dec 2026
Agent trial ends
"A free trial of all Data Cloud agents through December 31, 2026, after which the new pricing will become effective."
January 2027
First agent invoice
Every question asked of a Conversational Analytics agent from 1 January is billed in data tokens.
One FAQ, two dates
Google's own billing FAQ gives 27 Oct 2026 and 2 Nov 2027
The Dataplex FAQ, last updated 13 October 2026, says "Active billing for data insights starts on October 27, 2026", then in the next note that "the billing rollout for data insights begins on November 2, 2027 and takes up to a week to be enabled in all regions". One of those years is almost certainly a typo. Plan for the earlier date: Google says token metering is already switched on, so the usage is being counted either way.
The trial already moved once
The agent trial was set to end on 30 September
When we first read Google's Data Cloud agents page, the free trial ran to 30 September 2026. It now runs to 31 December. An extension is good news for a pilot, but it is also a reason not to build next year's budget on a free line item. Price the agents as if January is real.
›_ bigquery conversational analytics pricing per question
What a plain-English question costs in data tokens.
Google publishes the token rates but no per-question figure. These three rows are our assumptions, priced exactly at Google's rates, so you can see how the bill moves.
| Question shape (assumed) | Data tokens per question | Per question | 1,500 a month | 5,000 a month |
|---|---|---|---|---|
| Small schema, short answer | 10,000 in, 1,000 out | $0.05 | $75 | $250 |
| Mid-size model, chart and summary | 30,000 in, 1,500 out | $0.12 | $180 | $600 |
| Wide tables, follow-up questions | 60,000 in, 2,000 out | $0.22 | $330 | $1,100 |
The arithmetic for the middle row: 30,000 input tokens at $3 per million is 9 cents, 1,500 output tokens at $20 per million is 3 cents, so 12 cents a question. Input dominates, and input is mostly context: the table schemas, field descriptions, agent instructions and the earlier turns of the conversation that get sent with every follow-up. A wide dataset and a chatty user cost more than a narrow dataset and a precise one, even for the same answer.
None of this includes the query itself. Google's conversational analytics docs say you are "charged at BigQuery compute pricing for queries that run when you create data agents and have conversations". If each question scans 20 GB on demand, 1,500 questions scan about 29.3 TiB, which is roughly $177 a month after the free TiB. Partitioned, clustered tables cut that number far more than any token setting does.
How to get your real number
- Run the pilot now, while the agents are free, with the questions your team actually asks.
- Read the token metering Google provides. It says metering is available before active billing starts.
- Divide tokens by questions, then multiply by $3 and $20 per million for input and output.
- Pull the BigQuery jobs the agent ran from INFORMATION_SCHEMA and add the bytes billed.
›_ what google's docs say that changes the budget
Four lines in Google's documentation worth reading before January.
Quoted from the Dataplex billing FAQ, the conversational analytics overview and the Conversational Analytics API cost guide.
1. Service accounts pay from token one
"Service accounts don't receive a free per-user quota allowance." The wording implies people get a free per-user allowance and pipelines do not, so anything scheduled or embedded under a service account is fully billable once billing starts.
2. Spend caps only work on demand
Google's cost controls for agents, project and per-user daily quotas plus a big_query_max_billed_bytes limit, "apply only to projects that use on-demand billing and not to projects that use slot reservations". Its advice is to give each costly agent a project of its own.
3. No quota per agent
"Setting quotas for individual agents isn't supported." If marketing and finance each get an agent in the same project, you cannot cap one without capping the other.
4. Graph questions need an edition
When an agent answers with Graph Query Language, those queries "require a BigQuery Enterprise or Enterprise Plus edition reservation". On on-demand pricing, Google says they fail with an error.
To switch data insights off, Google says an administrator can disable the Dataplex API in the project, and the feature needs the Gemini Data Analytics API enabled to run in the first place. Our guide to BigQuery cost control for ad hoc queries covers the quota settings in more detail.
›_ bigquery data agent pricing vs the alternatives
Plain-English questions on each warehouse, priced side by side.
Every row bills the warehouse compute separately. The AI meter is the part that differs.
| Tool | AI meter | Free period | Query compute | Source and date |
|---|---|---|---|---|
| BigQuery Conversational Analytics Agent | Data tokens, $3 in and $20 out per million | Free trial to 31 Dec 2026 | BigQuery bytes or slots, billed separately | cloud.google.com, read 15 Oct 2026 |
| Snowflake Cortex Analyst | 67 credits per 1,000 messages, $0.134 a question on Standard | None published | Warehouse time, billed separately | Snowflake Credit Consumption Table, effective 12 Oct 2026, read 15 Oct 2026 |
| Databricks Genie | DBUs, $0.070 per DBU past 150 free DBUs a user | Free until 31 Jan 2027 | SQL warehouse DBUs, billed separately | Databricks pricing, read 8 Oct 2026 |
| AgentSQL Team | $59 a month billed yearly for 1,500 questions, about 3.9 cents each | Console demo on this page | BigQuery bytes, with a dry-run estimate and a cap per query | Our pricing page, read 15 Oct 2026 |
Google's meter is the hardest of the three cloud options to forecast, because it bills tokens rather than questions and publishes no per-question estimate. Snowflake's per-message rate is the most predictable, and Databricks has the longest free period. The details are on Snowflake Cortex Analyst pricing, Databricks Genie pricing and our conversational analytics pricing page, which puts every vendor's AI meter into one unit.
›_ a flat price for the metered half
Keep Gemini for your analysts and give everyone else a flat-priced way to ask.
If your analysts live in BigQuery Studio, keep Gemini's code assist switched on. It is included in what you already pay, and it is good at what it does. The harder purchase is the other half: letting the people who do not open BigQuery Studio ask their own questions, without a token bill that rises with every follow-up.
That is the job AgentSQL does. It connects to BigQuery with a read-only scope, turns a plain-English question into SQL, runs it, and returns a chart, a table and a one-line answer with the query shown so an analyst can check it. Team is $59 a month billed yearly for 1,500 questions and up to 10 seats, so the AI part of the bill is a known number in January and every month after.
BigQuery still bills the queries, and we treat that as your money. Every question is dry-run first so the estimated scan is known before it runs, every job carries a maximum bytes billed cap (100 GB by default, set per connection), and every job is labeled so you can find it in your billing export. If you also keep data in Postgres, MySQL or Snowflake, the same plan asks those too. Prices are on our pricing page.
The January math
$59
a month billed yearly for 1,500 questions on AgentSQL Team, about 3.9 cents each
$75 to $330
a month in data tokens for the same 1,500 questions on Google's agent, on our three assumed question shapes
Both figures exclude BigQuery compute, which is the same whichever tool writes the SQL.
›_ frequently asked
Gemini in BigQuery pricing questions, answered.
Is Gemini in BigQuery free?
Partly. Google says SQL code assist, Python code assist, data canvas and data preparation are available on every BigQuery edition and on demand at no additional cost. The agent features are not: Data Insights starts billing on 27 October 2026, and the Conversational Analytics, Data Science and Data Engineering agents are free only through 31 December 2026.
How much does Gemini in BigQuery cost?
The core coding features cost nothing extra. The Data Cloud agents, which include Conversational Analytics and Data Insights, cost $3 per million input data tokens and $20 per million output data tokens once their free period ends. Every query those features run is billed at normal BigQuery rates, $6.25 per TiB on demand after the first free TiB each month.
How much does BigQuery conversational analytics cost?
Nothing for the AI part until 31 December 2026, then $3 per million input and $20 per million output data tokens. Google publishes no per-question figure. On our token scenarios that is roughly 5 to 22 cents a question, plus the BigQuery compute for the SQL the agent runs, which is billed from day one.
When does Data Insights billing start in BigQuery?
Google's Dataplex billing FAQ says active billing for data insights starts on 27 October 2026. The same FAQ also says the rollout begins on 2 November 2027 and takes up to a week across regions, so one of the two dates is likely a typo. Budget from the earlier date, since metering is already available.
Do service accounts get free Gemini data insights?
No. Google's FAQ states that service accounts do not receive a free per-user quota allowance for data insights, although their usage is metered the same way as people. If you run scans or agents from a pipeline under a service account, every token is billable once billing starts.
How do I stop being charged for Gemini data insights?
Stop triggering new scans and let current ones finish. Google says an administrator can turn the capability off by disabling the Dataplex API in the project. Data insights also need the Gemini Data Analytics API enabled to run at all, so a project without it will not produce insights or charges.
Can I cap what a BigQuery data agent spends?
Only partly. Google lets you set project and per-user daily query quotas and a big_query_max_billed_bytes limit on an agent's queries, but its docs say these controls work only on on-demand projects, not slot reservations, and quotas for individual agents are not supported. Google's advice is to isolate each costly agent in its own project.
Is there a flat-priced alternative to BigQuery conversational analytics?
Yes. AgentSQL connects read-only to BigQuery and answers plain-English questions for a flat $59 a month billed yearly on Team, which includes 1,500 questions and up to 10 seats. BigQuery still bills the queries, but every one is dry-run first and carries a bytes-billed cap, 100 GB by default.
›_ bigquery and warehouse ai pricing
›_ connect your database