Google shipped close to 30 BigQuery updates between 1 and 27 October 2026. Most are about AI inside SQL and about "conversational analytics", BigQuery's natural-language data agents. A few are simple quality-of-life improvements that every SQL writer will notice. This roundup picks out what matters for analysts and data engineers, with what is generally available (GA) and what is still in preview. For the same view of other platforms, see our Snowflake and Databricks October updates.
SQL quality of life
Reusable column aliases
The most-requested change for many SQL writers: once enabled, an alias defined in the SELECT list can be used later in the same SELECT list and in the WHERE clause, instead of repeating the expression or wrapping the query in a CTE.
SELECT
net_price * 1.23 AS gross_price,
gross_price * 0.5 AS discounted_gross_price
FROM product
WHERE discounted_gross_price < 10.0;It is off by default. A BigQuery admin turns it on for a project or a whole organisation, per region:
ALTER PROJECT my_project
SET OPTIONS (
`region-eu.reusable_column_aliases_mode` = "on"
);- Copy tables across regions (GA)
- Copy a table from, for example, the US to the EU with a normal table copy, useful for data residency and migrations.
- REGEXP_EXTRACT_GROUPS (preview)
- Returns the text matched by several capturing groups in one call, instead of one REGEXP_EXTRACT per group.
- Flexible column names for Iceberg tables (GA)
- Iceberg external and managed tables now support flexible column names by default.
- Rust SDK (GA)
- The BigQuery client library for Rust is generally available.
AI functions in SQL
BigQuery keeps adding AI functions that run a model over your rows inside a query. October brought three to general availability and a new embedding model.
AI in BigQuery, October 2026
Gemini Embedding 2 (GA)
Embeds long, multilingual and multimodal content, including text, images, audio, video and PDFs, in AI.EMBED, AI.SIMILARITY and AI.GENERATE_EMBEDDING.
TimesFM 3.0 (preview)
Google's time series model now supports multivariate forecasting in AI.FORECAST, plus AI.EVALUATE and AI.DETECT_ANOMALIES.
AI.AGG (GA) and Gemini 4 Argon
Summarise many rows with a natural-language instruction; Gemini 4 Argon available in generative AI functions.
AI.CHUNK_DOCUMENT (GA)
Splits documents into coherent text chunks for embeddings and retrieval.
AI.TRANSLATE (GA)Now
Translates text between languages with standard machine translation or a translation LLM.
- AI.AGG
- An aggregate function: instead of SUM or COUNT, it reads all the text in a group, such as the reviews for each product, and returns a summary that follows your instruction. A new output_schema argument returns structured fields instead of free text.
- AI.CHUNK_DOCUMENT
- A table function that cuts long documents into pieces of the right size to embed, the first step of retrieval-augmented generation (RAG).
- AI.TRANSLATE
- Translates a text column, choosing between classic neural machine translation and a translation LLM.
- Gemini Embedding 2
- One embedding model for many kinds of content, so you can search images and PDFs with a text question.
- TimesFM 3.0
- A pre-trained forecasting model: no training step, and now able to use extra variables such as price or promotions.
Conversational analytics and agents
BigQuery's data agents answer questions in plain language. This month they became easier to govern and to share:
Sharing and reach
GA- Publish an agent to Gemini Enterprise (14 October)
- Conversations across several projects, or started from a dataset (14 October)
- Export answers, charts and tables to PDF or CSV (15 October)
Business users can get answers where they already work.
Governance
GA- Log every agent conversation to a BigQuery table (26 October)
- Chat with several graphs, tables, views and UDFs as sources (1 October)
- AI.CAUSAL_EFFECT for intervention analysis is still in preview
Logging lets you audit what was asked and check answers.
Data engineering and Lakehouse
- Data Engineering Agent can take Google Sheets, CSV and JSONL files from Google Drive as pipeline input (GA, 27 October), and now supports newer Gemini Flash models.
- Pipeline metadata enrichment and data quality scorecards are GA for BigQuery pipelines (1 October).
- Incremental transfers for PostgreSQL and ServiceNow in the Data Transfer Service (GA, 12 October).
- Lakehouse runtime catalog: a serverless Apache Hive endpoint (GA) and a migration path from Dataproc Metastore (GA, 14 October); change data capture from Delta Lake and Iceberg into Lakehouse Iceberg tables is in preview.
- Python UDFs can mount dependency files from Cloud Storage (preview, 26 October).
What to do this month
- 1
Decide on column aliases
Test reusable aliases in one project, check your scheduled queries still return the same results, then roll out.
- 2
Try AI.AGG on a small sample
Summarise customer feedback or support tickets by product, with LIMIT, and compare the summary to a manual read.
- 3
Revisit forecasts
If you forecast with ARIMA_PLUS, compare it against TimesFM 3.0, especially where promotions or prices drive demand.
- 4
Turn on conversation logging
If business users already use data agents, log conversations before usage grows.
Check your understanding
1/3What do reusable column aliases let you do?
They are off by default and must be enabled per project or organisation.
2/3What kind of function is AI.AGG?
AI.AGG reads the text in each group and follows a natural-language instruction.
3/3Which model adds multivariate forecasting to AI.FORECAST?
TimesFM 3.0 is in preview and can use extra variables alongside the series you forecast.
You got 0 out of 3.
More platform news: this week's data platform news, Snowflake October 2026 update, Databricks October 2026 update and how to choose a database.
Sources: Google Cloud BigQuery release notes, 1 to 27 October 2026, and the BigQuery documentation on reusable column aliases. Features roll out by region; check availability and pricing in your project.
Alessandro Ecclesie AgazziWritten by Alessandro Ecclesie Agazzi, analytics engineer in London. Updated 28 October 2026. About me