Gemini in BigQuery pricing: free SQL assist, $3 and $20 per million agent tokens
Gemini in BigQuery has two price levels, and the line between them decides most of the bill. The core assistance features (SQL code generation and explanation, Python code generation and completion, data canvas and data preparation) are listed by Google Cloud as available at no additional cost on every BigQuery compute option: Standard, Enterprise and Enterprise Plus editions and on-demand compute. The agent features are metered separately at $3 per million input data tokens and $20 per million output data tokens, under what Google calls Data Cloud Agent pricing.
The agent rate covers the Data Science Agent, the Data Engineering Agent, the Conversational Analytics Agent and data insights. Google Cloud lists a free trial of all Data Cloud agents through December 31, 2026, with the token pricing taking effect after the trial. Data insights is the exception with its own date: Google's Knowledge Catalog FAQ states that active billing for data insights started on October 27, 2026, with a rollout beginning November 2, 2026 that can take up to a week to reach all regions.
Neither level replaces normal BigQuery compute charges. Every query that Gemini writes and you run, and every query an agent runs on your behalf, is billed at BigQuery compute rates. The AI cost of Gemini in BigQuery is therefore zero for code assistance, token-based for agents, and in every case added to the query cost the team already pays.
··········
GEMINI IN BIGQUERY PRICING BY FEATURE.
The figures below are as listed on Google Cloud's Gemini for Google Cloud pricing page, Data Cloud Agent pricing page, BigQuery pricing page and the BigQuery and Knowledge Catalog documentation, all verified on November 14, 2026.
........
Feature | Status listed by Google | Gemini charge | Also billed |
SQL code generation and explanation | GA | No additional cost | Queries you run |
SQL code completion | Preview | No additional cost | Queries you run |
Python code generation and completion | GA | No additional cost | Queries you run |
Data canvas | GA | No additional cost | Queries you run |
Data preparation | GA | No additional cost | BigQuery jobs it runs |
Data insights | Billing active since Oct 27, 2026 | $3 / $20 per million input / output tokens | Generated queries you run |
Conversational Analytics Agent | Free trial to Dec 31, 2026 | $3 / $20 per million tokens after trial | BigQuery compute for its queries |
Data Science Agent | Free trial to Dec 31, 2026 | $3 / $20 per million tokens after trial | BigQuery jobs it runs |
Data Engineering Agent | Free trial to Dec 31, 2026 | $3 / $20 per million tokens after trial | BigQuery jobs it runs |
........
Google also offers custom quotes for agent usage through its sales team, so the $3 and $20 rates are the list pay-as-you-go prices.
The Data Cloud Agent pricing page gives both the October 27, 2026 start for data insights and the trial running to December 31, 2026 without reconciling them. The Knowledge Catalog FAQ is specific that data insights billing is active, so teams should treat data insights tokens as billable today and the three agents as billable from January 2027.
··········
BIGQUERY COMPUTE BEHIND GEMINI-GENERATED QUERIES.
The query cost is separate from the Gemini cost and follows the project's normal BigQuery pricing model. On-demand compute is listed at $6.25 per TiB in the US multi-region, with the first 1 TiB of query data processed per month free. Editions are billed in slot-hours on pay-as-you-go terms at $0.04 per slot-hour for Standard, $0.06 for Enterprise and $0.10 for Enterprise Plus in the US price tables, billed per second with a one-minute minimum. The full comparison of the two models, including commitments, is in our reference on BigQuery on-demand versus editions pricing.
The BigQuery conversational analytics documentation states that you are charged at BigQuery compute pricing for queries that run when you create data agents and when you hold conversations. The data insights documentation adds that you need sufficient permissions to run the generated queries. A free SQL suggestion that scans a 3 TiB table still costs the same as a query typed by hand, which is why our guide on writing SQL with Gemini in BigQuery treats the bytes-scanned estimate as the first check on any generated query.
Some agent queries also depend on the edition. The conversational analytics documentation notes that GQL graph queries require an Enterprise or Enterprise Plus reservation and fail under on-demand pricing, apart from some GRAPH_EXPAND queries. Agents cannot run DML, write operations or remote functions, so their compute is limited to reads.
··········
HOW GOOGLE COUNTS DATA AGENT TOKENS.
Google splits metered usage into input tokens, meaning the context, instructions and history sent to the model, and output tokens, meaning the generated descriptions, relationships and queries. The Knowledge Catalog page on monitoring data insights token usage defines the counts that appear in the logs for each successful scan job. Non-cached input tokens equal the prompt token count minus the cached content token count, plus tool-use prompt tokens. Cached input tokens are listed separately. Output tokens equal the candidates token count plus the thoughts token count, so the model's thinking tokens are billed as output.
Because output costs more than six times as much as input, long answers and extended reasoning raise the cost per question faster than large inputs do. Google does not publish a typical token count per question or per table scan.
Only jobs that reach a succeeded state and carry a billable token usage block are counted in the monitoring queries. Google states that service accounts receive no free per-user quota allowance and that their usage is metered like that of standard users. Data insights also requires the Gemini Data Analytics API (geminidataanalytics.googleapis.com) to be enabled, otherwise the scans fail.
··········
GEMINI IN BIGQUERY QUOTAS PER USER.
The no-cost features are limited by quotas rather than by price. The Gemini for Google Cloud quotas page, last updated October 28, 2026, lists these values for Gemini in BigQuery.
........
Quota | Value | Scope |
Requests per second | 2 | Per user per project, all Gemini for Google Cloud products |
Code generation and completion requests per day | 6,000 | Per user per project |
Chat, visualization, data insight scans, metadata generation, data preparation per day | 960 | Per user per project |
........
Data insights scans and metadata generation fall in the 960-per-day category. The setup documentation states that data insights and automated metadata generation are available to customers on on-demand compute, Enterprise edition or Enterprise Plus edition, and that their quotas are based on organization-level use of those compute models.
For data insights, the Knowledge Catalog FAQ describes daily input and output token quotas that are checked before each scan starts. A scan already running when the limit is crossed finishes, rejected scans incur no charge, and the counters reset at midnight Pacific Time. Administrators can raise the limit or set it to unlimited on the Quotas and System Limits page.
··········
ILLUSTRATIVE COST EXAMPLE FOR A 10-ANALYST BIGQUERY TEAM.
The example below is illustrative. The token counts per scan and per question and the bytes scanned are assumptions chosen for the calculation, since Google publishes no typical values; the prices are the list rates above. The team has 10 analysts in a US project on on-demand compute.
Ordinary querying, including SQL written with Gemini code assistance, scans 12 TiB a month. After the free 1 TiB, 11 TiB at $6.25 costs $68.75, and Gemini adds nothing for the suggestions. Data insights runs 40 table scans a month at an assumed 60,000 input and 8,000 output tokens per scan, which is $0.34 per scan and $13.60 a month. The Conversational Analytics Agent answers 100 questions per analyst per month, 1,000 in total, at an assumed 15,000 input and 1,500 output tokens per question, or $0.075 per question and $75.00 a month in tokens once the trial ends. Each question scans an assumed 2 GiB, so 2,000 GiB (about 1.95 TiB) adds $12.21 of query compute.
........
Cost line | Volume (illustrative) | Monthly cost |
Analyst queries incl. Gemini-written SQL | 12 TiB, 1 TiB free | $68.75 |
Gemini SQL and Python assistance | Unlimited within quotas | $0.00 |
Data insights tokens | 40 scans | $13.60 |
Conversational analytics queries | 1,000 questions, 2,000 GiB | $12.21 |
Conversational analytics tokens (from 2027) | 1,000 questions | $75.00 |
Total during the 2026 trial | $94.56 | |
Total from January 2027 | $169.56 |
........
On these assumptions the agent tokens become the second-largest line after ordinary querying, and the cost per analyst rises from about $9.46 to $16.96 a month when the trial ends. The usage stays far inside the quotas: 100 questions a month is roughly five a working day against a limit of 960 requests per user per day. Doubling the output tokens per question would add another $30.00 a month, which shows why answer length drives the agent bill more than question volume.
··········
CONTROLLING GEMINI IN BIGQUERY COSTS.
The Conversational Analytics API cost documentation recommends three controls for agent queries. The first is the big_query_max_billed_bytes option in the agent's data source settings: before running a query, the agent does a trial run to estimate bytes scanned, and a query above the limit fails without a charge. Google's example sets 100 MiB (104,857,600 bytes); in the illustrative team above, a cap would have to sit above the 2 GiB average or most questions would fail.
The second and third controls are the BigQuery custom quotas "Query usage per day" at project level, which Google describes as a hard stop on daily cost, and "Query usage per day per user". Google recommends a dedicated project for each agent before applying them, because per-agent quotas and per-agent cost tracking with BigQuery labels are not supported. These byte-based controls apply only to on-demand projects, not to projects running on slot reservations, where per-project limits such as the newly GA BigQuery project caps are the relevant tool.
For data insights, the token quotas described above act as the spend limit, and the Knowledge Catalog FAQ adds that stopping ongoing scans and new triggers, or disabling the Dataplex API in the project, stops the charges. Token usage per scan job can be read in Logs Explorer or in a Cloud Monitoring dashboard built from Google's published definition, which Google says estimates costs and should not be used to reconcile an invoice.
Access control is the last lever. The setup documentation lists the BigQuery Studio User and BigQuery Studio Admin roles and the cloudaicompanion permissions behind Gemini in BigQuery; revoking those permissions removes a user's access, individual features can be cleared in the Gemini settings of the BigQuery toolbar, and turning off the Gemini for Google Cloud API disables all Gemini for Google Cloud products in the project.
··········
GEMINI IN BIGQUERY: COMPLIANCE AND WHAT TO CHECK BEFORE ROLLOUT.
Google's setup page states that Gemini in BigQuery is part of Gemini for Google Cloud and does not support the same compliance and security offerings as BigQuery, and it advises enabling it only in projects that do not need the offerings Gemini lacks. The Data Science Agent documentation adds that customer-managed encryption keys are not supported for that agent and that its code runs only in the notebook runtime. Teams with regulated data should read the separate data-processing locations page before enabling any feature, and the same applies to agents used from Google Sheets, covered in our story on conversational analytics in Connected Sheets.
··········
GEMINI IN BIGQUERY COSTS: WHAT GOOGLE CONFIRMS AND WHAT DEPENDS ON YOUR PROJECT.
Google Cloud confirms that SQL and Python code assistance, data canvas and data preparation carry no additional Gemini charge on any BigQuery compute option, that data insights and the Data Science, Data Engineering and Conversational Analytics agents are priced at $3 per million input tokens and $20 per million output tokens, that data insights billing has been active since October 27, 2026, and that the agent trial runs through December 31, 2026. Every query generated or run by Gemini is billed at the project's BigQuery compute rate.
What depends on the project is the size of each line: whether it runs on-demand or on an edition reservation, how many bytes generated queries scan, how long agent answers are, and whether quotas and byte caps are set. Google does not publish typical tokens per question or per scan, so the first month of logs is the only reliable basis for a 2027 budget, and the cost of a question asked of an agent is the sum of its tokens and the query it runs.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
