Home » AI Models Guide » Multiple Models

How to Use Multiple AI Models in One Workflow

Using multiple AI models in a single workflow lets you optimize each step for what matters most: cheap models for simple routing, mid-tier models for standard processing, reasoning models for complex analysis, and premium models for customer-facing output. This mixed-model approach typically saves 40 to 70% compared to using a premium model for every step, while actually improving overall quality.

Why One Model Does Not Fit All Steps

A typical business workflow includes steps with very different requirements. Classifying an incoming message is a simple task that any model handles well. Analyzing data requires accuracy. Writing a customer response requires quality. Using the same expensive model for all three steps wastes money on the simple step and might not even be the best choice for the analysis step (where a reasoning model outperforms a chat model).

The Mixed-Model Pattern

The platform's workflow automation and chain commands let you configure a different model at each step. Here is a common pattern:

Step 1: Intake and Classification (GPT-4.1-nano)
Classify the incoming message, detect intent, extract key fields. This step runs on the cheapest model because classification accuracy is high even on nano models. Cost: under 1 credit.
Step 2: Data Retrieval (No model needed)
Query databases, pull customer records, retrieve knowledge base results. This step does not use an AI model at all, just database queries.
Step 3: Analysis or Decision (GPT o3-mini)
If the workflow requires calculation, comparison, or complex decision-making, route to a reasoning model. It costs more per request but gets the answer right. Cost: 5 to 15 credits.
Step 4: Response Generation (Claude Sonnet)
Generate the customer-facing response using a writing-optimized model. The output is what the customer sees, so quality matters here. Cost: 3 to 5 credits.

Common Multi-Model Setups

Customer Support Pipeline

Nano classifies the ticket type, mini retrieves and formats relevant knowledge base information, Sonnet drafts the response. Total cost per ticket: 4 to 7 credits instead of 10 to 15 using a premium model for everything.

Data Analysis Report

Mini formats and cleans the raw data, o3-mini analyzes patterns and calculates metrics, Opus writes the final report. The reasoning model ensures accurate calculations while Opus ensures readable output.

Lead Qualification

Nano extracts contact information from the form submission, mini scores the lead based on criteria, Sonnet generates a personalized follow-up message. See How to Qualify Leads With AI.

Content Moderation

Nano performs initial content screening (fast, cheap), and only flagged content gets escalated to a mid-tier model for nuanced review. This reduces costs by 80 to 90% compared to running every piece of content through a premium model.

How to Set This Up

In the visual workflow builder, each step has a model selector. Choose the appropriate model for each step's role. You can also set different models in custom app code by specifying the model parameter in each AI function call.

Start by building the workflow with a single mid-tier model for all steps, then optimize by swapping individual steps to cheaper or more specialized models. Test the output quality after each change to make sure accuracy remains acceptable.

Build smarter workflows with the right model at each step. Start building and optimizing.

Get Started Free