Cortex Analyst vs Databricks Genie Pricing and the Real Cost per Question

Snowflake Cortex Analyst costs 67 Platform Credits per 1,000 messages, which is $0.134 a question on Standard, $0.201 on Enterprise and $0.268 on Business Critical at AWS US East on-demand prices. Databricks Genie is free for human users until 31 January 2027, then bills $0.070 per DBU after 150 free DBUs per user each month, and Databricks publishes no DBU figure per question. On both platforms the warehouse that runs the generated SQL is billed on top, and on Databricks that warehouse (SQL Serverless at $0.70 per DBU) costs ten times the Genie rate per DBU.

So the honest comparison is not one price against another. It is a price you can forecast against one you have to measure. The Snowflake figures below come from Snowflake's Credit Consumption Table marked effective 12 October 2026, and the Databricks figures from its Genie and Databricks SQL pricing pages and Genie release notes, all read on 15 October 2026. Where a number is derived, the arithmetic is shown.

The two rate cards side by side

Cortex Analyst turns a plain-English question into SQL against a semantic model or semantic view that someone on your team writes. Genie does the same job inside a Genie Agent (the setup Databricks used to call a Genie space), which sits on Unity Catalog tables and needs a pro or serverless SQL warehouse to run. Genie One is the chat interface business users see.

One Snowflake detail changes the per-question figure. The message price is in Platform Credits, so the edition and region set the dollar amount ($4.00 a credit on Business Critical, $3.90 for Enterprise in EU Dublin), and a negotiated capacity discount applies to it. If your team reaches Cortex Analyst through Cortex Agents instead, the table says the per-message price does not apply and model tokens bill in AI Credits at $2.00 on demand ($2.20 pinned to your region), which the Platform Credit discount does not reduce. Our Snowflake Cortex Analyst pricing breakdown covers both routes in detail.

What 1,000 and 5,000 questions a month cost on each

Assume each person asks 100 questions a month, roughly five a working day. That puts 1,000 questions at a 10 person team and 5,000 at a 50 person team. Snowflake is on demand in AWS US East through the API route; Databricks is Premium on AWS US East.

Cortex Analyst at 1,000 and 5,000 questions

• 1,000 questions times 0.067 credits is 67 credits. At $2.00, $3.00 and $4.00 a credit that is $134, $201 or $268 a month on Standard, Enterprise or Business Critical.
• 5,000 questions times 0.067 credits is 335 credits, so $670, $1,005 or $1,340 a month.

Then add the warehouse. As an illustration only, an XS warehouse awake four hours each working day is 4 times 21, or 84 credits a month: $168 on Standard, $252 on Enterprise. Your number depends on how your questions cluster through the day, because each resume after suspend bills at least one minute.

Genie at 1,000 and 5,000 questions

• Until 31 January 2027 the Genie line is $0 at either volume for human users.
• From 1 February 2027, 10 users get 10 times 150, or 1,500 free DBUs, worth $105. At 1,000 questions the Genie line stays at $0 as long as the average question uses no more than 1.5 DBUs (1,500 divided by 1,000).
• 50 users get 7,500 free DBUs, worth $525. At 5,000 questions the threshold is again 1.5 DBUs a question (7,500 divided by 5,000).
• Above the allowance, every extra DBU is $0.070. A Genie question would need about 2.9 billable DBUs to cost what an Enterprise Cortex Analyst message costs ($0.201 divided by $0.070).

Then add the SQL warehouse. Databricks shows each warehouse's DBUs per hour in its settings; multiply that by the hours it stays awake and by $0.70 on Serverless. Serverless warehouses stop after 10 idle minutes by default (5 is the lowest the UI allows, 1 through the API), and Pro warehouses default to 45 minutes with a 10 minute minimum. That idle tail is where an interactive question workload spends money on both platforms. The Databricks pricing guide has the full SQL rate card and what the serverless rate includes.

The embedded case

If the questions come from an application authenticating as a service principal rather than from people, Genie has no allowance and no promotion. Databricks' release notes state that since 24 September 2026, service principal usage bills at standard pay-as-you-go pricing. At 5,000 questions a month, every 1 DBU of average consumption per question is 5,000 times $0.070, or $350 a month. Cortex Analyst prices the same 5,000 embedded questions exactly as it prices human ones.

Forecasting risk, per message versus per DBU

A finance team can sign off on the Cortex Analyst number before anyone asks a question. Volume times 0.067 credits times your credit price is the whole formula, and the CORTEX_ANALYST_USAGE_HISTORY view lets you check it afterwards. The risk on Snowflake is not the unit price. It is the route: if your developers move from the Analyst API to Cortex Agents, which Snowflake's documentation recommends for answer quality, the bill switches to tokens in AI Credits, and the cost per question then depends on model choice, semantic model size and how many steps the agent takes.

Genie's risk sits in the other place. The unit price is clear and low, but the quantity is unknown until you run your own questions. A simple count over one table and a multi-step exploration across several joined tables are both one question to the user and very different amounts of work to the meter. The allowance makes that tolerable for people: a user who exceeds 150 DBUs is the exception, and Databricks' own pricing page says many users never exceed it. It does not help with the warehouse, which is the larger number, and it does not exist for service principals.

The practical plan for Genie is to use the free months as a measurement period. Pull Genie usage from the system.billing.usage table per user each week until January, and you will have the per-question DBU figure that Databricks does not publish, from your own workload, before the first invoice arrives. Our Databricks Genie pricing article covers the allowance and the service principal exclusion in more detail.

Where each one wins

Cortex Analyst wins when the budget has to be fixed in advance, when questions will come from an application, and when your team already maintains metric definitions that translate cleanly into a semantic view. A governed semantic model also means two people asking for revenue get the same definition, which is worth more than the per-message price to some finance teams.

Genie wins when your data already lives in Unity Catalog and your users already have workspace access. Until February it costs nothing on the Genie line, and afterwards Databricks says many users never exceed their 150 DBU allowance. The catch is the warehouse tier: a team on SQL Classic has to move to Pro or Serverless to use Genie, and that move can cost more than Genie itself.

If you are still choosing the platform rather than the feature, do not let either assistant decide it. The warehouse is usually the larger line on either platform, and our Snowflake vs Databricks comparison and Snowflake pricing guide price that part properly. It also pays to watch the warehouse half of the bill alongside the rest of your cloud and SaaS spend from the first month, because neither assistant's usage view shows the compute it triggers. BigQuery teams face a third meter again, priced in our Gemini in BigQuery pricing breakdown, and the conversational analytics comparison lines up all the major vendors' units in one table.

The flat-priced option for Snowflake teams

If your warehouse is Snowflake and most questions are counts, sums and trends over tables that are already clean, a per-message meter is an expensive way to answer them. We built AgentSQL for that job, so weigh this section with that in mind. AgentSQL connects to Snowflake with a read-only role, reads the live schema, writes Snowflake SQL, runs it and returns the number, a chart and the query it wrote. There is no semantic model to write before the first question.

The Team plan is $59 a month billed yearly for up to 10 seats, 1,500 questions a month, and Snowflake and BigQuery connections. At the 1,000 question volume above, that is about 6 cents a question against $0.201 on Cortex Analyst Enterprise; used to its full 1,500 questions it is about 4 cents. The 50 person, 5,000 question team fits the Scale plan at $179 a month billed yearly, against $1,005 for the same volume on Enterprise Cortex Analyst. Starter at $24 connects Postgres and MySQL only.

Two limits to state plainly. Your Snowflake warehouse still runs every query AgentSQL writes, so the warehouse line in the examples above is the same whichever tool writes the SQL. And AgentSQL connects to Postgres, MySQL, Snowflake and BigQuery only. It does not connect to Databricks, so a Databricks team's choice is Genie or a tool built for its own platform. Read more about the Snowflake connection on AgentSQL for Snowflake, or type a question into the console at the top of this page and read the SQL it writes.