Google Cloud made BigQuery generative-AI token quotas generally available, enabling daily project-level and per-user limits aggregated across all regions.

Google’s October 14 article details recently announced general availability of token-based cost controls. BigQuery’s generative-AI functions bring large language models into SQL queries and are billed by token usage. Four daily quotas, managed through Google Cloud’s Identity and Access Management, or IAM, quota system, separately govern project-wide and per-user consumption.

The default limits distinguish input tokens from output tokens. Per project, the daily defaults are 200 billion input tokens and 20 billion output tokens; per user within a project, they are 150 billion input tokens and 15 billion output tokens. Usage is summed across all regions, output quotas include reasoning or “thinking” tokens, and per-user limits do not extend across an entire organization.

Coverage applies to the listed model-backed generation functions. These are AI.GENERATE, AI.GENERATE_TABLE, AI.GENERATE_TEXT and its ML.GENERATE_TEXT alias, AI.CLASSIFY, AI.IF, AI.SCORE, and AI.AGG. Embedding functions—including AI.GENERATE_EMBEDDING, ML.GENERATE_EMBEDDING, and AI.EMBED—are excluded because, according to Google, they use different and substantially cheaper models.

Administrators can manage the quotas through the console. The specified route is IAM & Admin → Quotas & System Limits, filtered to the BigQuery API; administrators can then select and edit the relevant quota. Lowering a quota takes effect immediately without approval, while the supplied excerpt directs administrators seeking higher usage to request an increase.