New Pipefy AI Models and Pricing

📍 Feature under release/research.

Pipefy AI Credits

Pipefy AI Agents consume AI credits. Each contracted plan includes 1,000 one-time AI credits. If needed, the customer can purchase additional packages of 500, 1,000, 10,000, or 50,000 credits.

New Pipefy AI models available to Agents

📢 Learn bout the phased rollout:

From July 13th onwards: the default Pipefy AI model will now be called Classic, and BYOM models remain unchanged. Additionally, three new models become available: Lite, Standard, and Pro.
- The Lite model consumes 2 credits
- The Standard model consumes 2 credits
- The Pro model consumes 3 credits (consumes 2 credits as promotional offer until October 8th)
From September 30 onwards: the Classic model is discontinued. All Agents using the Classic model automatically switch to the Standard model, which also consumes 2 AI credits.
From October 9, 2026 onwards: the Pro model promotional test period ends and it will consume 3 AI credits as of that date.

AI Credits consumption

AI credits are consumed according to the AI model chosen by the customer. See below how model selection works within a Pipefy AI Agent.

When creating and configuring a Pipefy AI Agent, the customer can define 1 to 5 different behaviors. Each behavior is composed of three fundamental definitions:

The behavior trigger: A trigger is any event partway through a process that activates the behavior.

Example: the card reached a certain phase; a file was attached to the card; an approval was performed; etc.

The behavior instructions: All sequential actions that the Pipefy AI Agent should perform after the trigger occurs are described in natural language.

Example: review a document; extract data from a form and save it to a database; send a notification to a user; etc.

Which AI model will be used by the Pipefy AI Agent to execute the instructions described in the behavior. Each of the three available models for selection consume different amounts of AI credits:

AI Model	Consumption
`Lite`	2 credits
`Standard` (default)	2 credits
`Pro`	3 credits (consumes 2 credits as promotional offer until October 8th 2026. On October 9th 2026, Pro model will consume 3 credits)

Three fundamental points:

The AI model is selected for each behavior, NOT for each Pipefy AI Agent.
Each Pipefy AI Agent can have up to 5 behaviors, which means a single Agent can function using more than one available AI model.
The choice of AI model for each behavior is 100% up to the customer.

Exceptions that don't consume AI Credits

Some exceptions in behaviors executed by Pipefy AI agents do not consume AI credits. They are:

If the automation is active but the condition for the Agent is not met → no consumption.
If the Agent receives data beyond the supported limit (such as more than 30 attachments per execution) → no consumption.
If there is a partial error (for example, a field with an invalid format, but other fields processed successfully), credit consumption will occur normally since part of the execution was completed.

Optimizing AI Credits consumption

The AI models are available so the customer can optimize credit consumption according to each type of behavior. Simpler behaviors can dispense with powerful models, reducing credit consumption and allowing the customer to select the best models only for the most complex behaviors.

The customer has freedom and control to choose and change the AI model for each behavior as many times as they wish.

The table below exemplifies the relationship between the three available AI models and typical actions found in Pipefy AI Agent behaviors:

AI Model	Ideal for
`Lite`	Ideal for: Short and predictable tasks, with clear inputs and a single result. Low reasoning depth, minimal ambiguity. When to choose: The instruction is simple, the data is already structured, and you expect 0 to 1 skill and 1 to 2 actions. Use cases: Language and tone: Translate card fields, rewrite a comment in another language, adjust tone (formal/informal). Simple content generation: Draft a short email, a Slack message, or the title/summary of a card from a few known fields. Field mapping and formatting: Copy values between fields, normalize dates/phone numbers, apply a fixed template. Simple classification: Mark a request (urgent/normal), choose a label from a small fixed list, direct to a phase when rules are obvious. Single-skill query: A Knowledge Base excerpt or a data query, followed by field update (ex.: "search for SLA from product code → set due date"). Confirmations and notifications: Send a standard email when a card enters a phase, with placeholders filled from card data. Examples in Pipefy: "When a card is created, write a welcome email in the requestor's language." "Read the Product Code, query the price in a table skill, fill the Estimated Value." Avoid when: there are multiple documents, nested conditionals, cross-field validation, or multi-step agentic planning.
`Standard`	Ideal for: For medium complexity tasks. Day-to-day operational automation — the default level for most behaviors. When to choose: You need to read documents or multiple fields, apply non-trivial logic, chain skills, and execute multiple actions on the card reliably. Use cases: Document extraction: Read invoices, contracts, forms, or PDFs and fill card/table fields (vendor, amount, dates, line items). Conditional flows: Evaluate business rules across multiple fields ("if amount > X and region = Y and category ∉ exceptions → move to Legal"). Agentic executions with multiple skills: Combine KB search + data query and then decide which actions to take. Examples in Pipefy: "On a new purchase card: extract the invoice PDF → validate against PO → update fields → move to 'Awaiting Approval' or 'Exception'." "Search KB for return policy → if within timeframe, update card and send standard email; otherwise, direct to supervisor phase." Sweet spot: 1 to 3 documents, multiple conditionals, multiple skills, multiple actions — typical process automation.
`Pro`	Ideal for: High-risk, unstructured, or analytically intensive work where errors are costly and reasoning quality matters more than speed. When to choose: Unstructured or conflicting sources, consolidation of multiple entities, nuanced judgment, or complex analysis over incomplete data. Use cases: Deep consolidation: Merge overlapping facts from multiple sources, resolve conflicts, and generate field values that are structured and audit-ready. Multi-step reasoning chains: long dependency chains ("extract → cross-reference with 3 policies → infer exceptions → propose an action plan → execute with guardrails"). Examples in Pipefy: "Analyze quarterly incident reports across regions, group root causes, recommend process changes, and create improvement cards with assigned owners." "Compare three quote PDFs against RFP requirements, score each vendor, document gaps, and move the card to the correct approval flow with a written justification." Use when: Quality, nuance, and accuracy are more important than latency and cost.

Using your own model (BYOM)

If you connect your own AI model to Pipefy, credit consumption works differently.

When using your own model via BYOM (Bring Your Own Models), each Agent run and Automation run will consume 1.5 credits.

BYOM	Additional Credit
Agent/Automation Run	+1.5 credits

Specific agentic capabilities such as document analysis, IDP, calculations, Web searches have additional charges that are explained below.

Additional AI Credit consumption

Some specific actions, when executed, consume additional AI credits:

Specific Action	Additional Credit
Document Analysis	+1 credit
Calculations or Analysis	+0.5 credit
Web Search	+0.5 credit

Examples:

In a behavior, the Pipefy AI Agent was configured by the customer to use the Standard AI Model, and one of its actions involves a web search. In this case, when the Agent executes the instructions of this behavior, it will consume 2 credits for using the Standard AI Model, plus 0.5 credit for performing the web search, totaling 2.5 credits consumed.

When analyzing an invoice, the Pipefy AI Agent executes a behavior with the Pro AI Model configured by the customer. Thus, total consumption will be 4 credits (3 credits for using the Pro AI Model, and 1 credit for the specific document analysis action).

Where to track consumption

The customer can track their organization's AI credit consumption in the Admin Panel.

Frequently Asked Questions

What are the AI Models available for Pipefy AI Agents?

Until July 12th, 2026, Pipefy AI Agents used a single AI Model: Classic. From July 13th, 2026 onwards, three new Models are available: Lite, Standard, and Pro.

How long will the Classic AI Model exist?

Until September 30th, 2026, the date when the Classic AI Model will be permanently discontinued.

How do the new AI Models work?

The new AI Models serve tasks with different complexity levels. Faster tasks can be executed by the Lite Model; medium-complexity by the Standard Model, and high-complexity by the Pro Model.

Which of the new AI Models corresponds to the Classic Model?

In terms of capability, the Classic Model corresponds to the new Standard Model.

What happens if I don't switch to one of the new AI Models?

On September 30, 2026, all behaviors of Pipefy AI Agents configured to use the Classic AI Model will automatically migrate to the Standard Model.

Why did Pipefy create new AI Models?

To give customers more freedom and control so they can optimize their AI Agents, according to their execution and tasks complexity.

How many credits do Pipefy's new AI Models consume?

The Lite AI Model consumes 2 credit per execution, regardless of how many actions are configured in its behavior. The Standard Model also consumes 2 credits and the Pro Model consumes 3 credits.

Until October 8, 2026, as a promotional offer, the Pro Model will consume only 2 credits to allow customers to test without cost impact.

Where do I select the new AI Models?

In the behavior configuration screen of the Pipefy AI Agent. Remember that each Agent can have up to 5 configured behaviors, each with its own chosen AI Model.