Snowflake Intelligence Pricing in AI Credits, Tokens and Real Monthly Math

Snowflake Intelligence has no seat fee. It bills the tokens its agent processes in AI Credits, priced at $2.00 each on demand with global routing ($2.20 with regional routing), at a per-model rate such as 1.95 AI Credits per million input tokens and 9.76 per million output tokens for claude-sonnet-4-6. Those rates come from the Snowflake Service Consumption Table effective 15 October 2026, and the warehouse that runs the generated SQL bills Platform Credits on top.

One naming note before the numbers. Snowflake's documentation now calls the product Snowflake CoWork. The Snowflake Intelligence pages redirect to CoWork pages, the usage view is SNOWFLAKE_COWORK_USAGE_HISTORY, and the consumption table prices it in Table 6(d), titled "Snowflake CoWork, Cortex Agents, & Cortex Analyst via Snowflake CoWork or Cortex Agents". Same product, same meter. Every figure below comes from that table and from Snowflake's AI pricing documentation, read on 19 October 2026. Where a number is our estimate, the token assumptions and the arithmetic are shown.

How much does Snowflake Intelligence cost?

About 12 cents a question in AI Credits, by our estimate, if each question uses 20,000 input and 2,000 output tokens on claude-sonnet-4-6. Warehouse compute comes on top. Snowflake publishes rates per million tokens, never per question, so your real figure depends on the model, the size of the prompt the agent builds and how many tools it calls.

Snowflake's AI pricing page describes the bill plainly: costs "are additive across the underlying services the agent invokes (such as Cortex Analyst and Cortex Search)", and "the specific rate depends on the model selected for orchestration." The same page says there are no per-seat fees. So a budget has three parts: orchestration tokens, tool usage, and the virtual warehouse that executes the SQL.

The AI Credit rates per million tokens

Table 6(d) lists close to 50 models. The ones below carry no Preview footnote in the 15 October 2026 table. The dollar columns multiply the credit rate by the $2.00 on-demand AI Credit price.

Three details in the table matter for a budget. Many newer models are marked Preview. Some carry promotional pricing with a written end: rates for gemini-3.7-flash and gemini-3.8-flash "will increase by 100% on January 1, 2027", gemini-4-argon doubles on 1 December 2026, and openai-gpt-5.6-sol "may increase by 25% on input and 50% on output on November 21, 2026." And Snowflake recommends the auto model setting, which "automatically selects the highest-quality model for your account", so the rate behind your agent can change without anyone editing it. Pin a model if you need a stable forecast.

Is Snowflake Intelligence included in my Snowflake contract?

There is no separate license to buy, but usage is not included. Snowflake says AI Credit pricing needs no contract amendment, and its own worked example adds AI Credit spend to Platform Credit spend for one total. Your negotiated Platform Credit discount does not apply to it. Capacity customers get their own AI Credit tiers by annual contract value, from $2.00 below $1.2 million down to $1.88 above $40 million.

AI Credits also cost the same on every edition. A Business Critical account pays $4.00 per Platform Credit in AWS US East (Standard $2.00, Enterprise $3.00), yet still $2.00 per AI Credit. The warehouse part of the bill follows your edition and region, which our Snowflake pricing guide covers in full.

Does Cortex Analyst cost extra inside Snowflake Intelligence?

Yes, but not at the per-message rate. The table lists Cortex Analyst at 67 Platform Credits per 1,000 messages, with footnote 25: "This pricing will only be applicable when using the Cortex Analyst API." Inside Snowflake Intelligence, Cortex Analyst runs as a tool, and its tokens bill in AI Credits at the Table 6(d) model rates.

The usage view shows this split, with token counts reported separately for the cortex_agents and cortex_analyst service types. For comparison, the API route costs $0.201 a message on Enterprise at $3.00 a credit. Our Cortex Analyst pricing breakdown covers both routes, and the Cortex Analyst and Databricks Genie cost comparison sets it against the Databricks meter.

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

These are estimates, built from stated assumptions so you can swap in your own:

• Orchestration uses 12,000 input and 1,200 output tokens per question (agent instructions, tool descriptions, conversation history and the reply).
• One Cortex Analyst tool call per question uses 8,000 input and 800 output tokens, priced at the same claude-sonnet-4-6 rate for simplicity.
• No prompt caching discount, global routing at $2.00 per AI Credit.
• An XS warehouse (1 credit an hour) on Enterprise at $3.00, awake 2 hours a working day at 1,000 questions and 8 hours at 5,000, over 21 working days.

The model choice moves the token line more than anything else. The same 20,000 input and 2,000 output tokens cost $0.039 a question on claude-haiku-4-5 and $0.195 on claude-opus-4-6, so 5,000 questions range from $195.00 to $975.20 in AI Credits. Regional routing adds 10 percent. Prompt caching pulls the other way, since a cache read on claude-sonnet-4-6 bills 0.20 AI Credits per million against 1.95 for fresh input.

Cortex Search is a separate, fixed line if your agent searches documents: 6.3 AI Credits per GB a month of indexed data, so a 10 GB index is 63 AI Credits, $126 a month, whether anyone asks a question or not. Deep research mode, code execution and web search tools add their own costs on top of orchestration, and Snowflake notes its list of cost sources "isn't exhaustive."

Why older cost guides show different rates

Some guides still price Snowflake Intelligence with superseded figures, such as claude-sonnet-4-5 at 3.45 credits per million input tokens multiplied by an edition's Platform Credit price, or Cortex Search at 6.3 Platform Credits per GB. The table effective 15 October 2026 lists claude-sonnet-4-5 at 1.95 AI Credits per million input tokens and 9.76 per million output, and Cortex Search at 6.3 AI Credits per GB a month, all priced at $2.00 per AI Credit regardless of edition. Edition credit prices quoted at around $2.50 for Enterprise also do not match Table 2(a), which shows $3.00 in AWS US East. Check the effective date on page one of whatever table a guide quotes.

How do I monitor Snowflake Intelligence spend?

Query the SNOWFLAKE_COWORK_USAGE_HISTORY view in ACCOUNT_USAGE; the documentation page for the older SNOWFLAKE_INTELLIGENCE_USAGE_HISTORY name now redirects there. Each row is one agent call, with TOKEN_CREDITS, a token and credit breakdown by service and model, the user, the agent, and a METADATA field holding the warehouse credits the request's SQL consumed.

Two caveats from the documentation. The sql_query_credits field can lag the rest of the row by up to 8 hours, and it excludes queries run on Adaptive Warehouses. Requests made through Cortex Agents directly land in CORTEX_AGENT_USAGE_HISTORY instead, so sum both views if you run both.

How do I cap Snowflake Intelligence costs?

Use three controls Snowflake documents. A resource budget on the Snowflake Intelligence object alerts or acts, for example by revoking access, when total spend crosses a threshold. Per-user quotas set a credit limit per person and can block users automatically until the cycle resets. And an orchestration budget on the agent limits what a single run can consume.

Caps on AI Credits do not cover the warehouse. Put the warehouse behind a resource monitor too, keep auto-suspend short, and give Snowflake Intelligence its own warehouse so its compute is easy to read. Cost is only half of the rollout risk. An agent answering from stale tables gives confident wrong numbers, so it is worth watching freshness and schema changes on the tables it queries before business users start relying on the answers.

The flat-priced option for Snowflake teams

If most of your users' questions are counts, sums and trends over tables that are already clean, a token meter with model-dependent rates is a hard thing to forecast. 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 a one-line answer, a chart, a table and the SQL it wrote. There is no semantic model to write before the first question.

The Team plan is $59 a month billed yearly ($118 billed monthly) for up to 10 seats and 1,500 questions a month, with Snowflake and BigQuery connections. Used to its full 1,500 questions, that is about 4 cents a question. The same 1,500 questions under the estimate above come to 87.78 AI Credits, $175.56, before warehouse. The 5,000 question team fits the Scale plan at $179 a month billed yearly, against an estimated $585.20 in AI Credits. Snowflake needs Team or above. For how other AI SQL tools charge per team, see our BlazeSQL pricing comparison, and all AgentSQL plans are on one page.

Two limits to state plainly. Your Snowflake warehouse still runs every query AgentSQL writes, so the warehouse line in the table above stays whichever tool writes the SQL. And AgentSQL connects to Postgres, MySQL, Snowflake and BigQuery only, and answers questions with SQL. It does not search PDFs, generate slide decks or run deep research, so teams that need those keep Snowflake Intelligence for them. The conversational analytics comparison lines up the other vendors' meters, or type a question into the console at the top of this page and read the SQL it writes.