The Divi AI Agent isn't tied to a single AI model. Every request it handles runs on a model you choose, and that choice is a trade-off between speed, quality, and cost
Important Note: A more capable model tends to think more carefully and handle harder work, but it also costs more per request and can take longer to respond.
The Automatic Options
By default, the Agent picks a model for you from one of four automatic tiers. You never have to think about this unless you want to - Auto Value is what you get out of the box.
Option | What it optimizes for | Suits |
Auto Value (default) | The lowest cost per request | Everyday requests - simple edits, content tweaks, small additions |
Auto Balanced | A middle ground between cost and capability | Requests that touch several settings or a few steps at once |
Auto Deep Value | More careful reasoning, still at a moderate price | Requests worth thinking through carefully, without paying premium rates |
Auto Premium | The most capable models available, regardless of price | Demanding work where getting it right the first time matters most |
Note: The exact models behind each tier can change over time. What stays true is the shape of the trade-off: moving from Auto Value toward Auto Premium buys you a more capable model at a higher cost per request.
Choosing a Specific Model
If you'd rather pick a model yourself instead of leaving it on automatic, hover over the model button in the AI Chat window (the message box at the bottom of the chat window).
The picker is a flat list containing the four automatic tiers and the Custom option. You can also type to search the list instead of scrolling it.
Models available through the Agent are served through OpenRouter, which is how Divi is able to offer a range of AI models instead of locking you into one provider.
Reasoning Effort
Alongside the model itself, you can set how much reasoning effort it puts into a request: Low, Medium (the default), High, or XHigh.
Raising it gives the model more room to think through a request before answering, which can help with anything that has several moving parts or an ambiguous instruction - at the cost of a slower response and, typically, a higher price per request.
Not every model on the list offers every level, so the options you see can vary depending on which model is selected.
Note: Leave this on Medium unless you have a specific reason to change it. It's a reasonable middle ground for most requests, and it's not a setting most people need to touch often.
Custom: A Different Model per Job
The Agent doesn't use just one model behind the scenes, even when you leave everything on automatic - it hands off different parts of the work to different internal roles. Selecting Custom from the model picker lets you choose the model for each of those roles yourself, rather than leaving it to the automatic tier's own choices.
Open the Model selection dropdown.
Click the Configure Custom Models button to open the Custom Model Selection window.
Choose the model and its reasoning level for each of the four roles.
Note: For each Role, the model selection dropdowns are grouped into two categories: Popular Models (a list of commonly used models) and All Models (a list of all avaiable AI models).
The four roles are:
Planner - works out the steps needed to carry out your request before any of them run. A stronger model here means a more reliable plan for complex jobs.
Agent - carries out the plan, step by step, actually making the changes on your page.
Layout - generates full page and section layouts, so the model you choose here has a direct effect on how good those layouts turn out.
Ask - answers your questions when you're using Ask mode rather than making changes. Since it's not doing hands-on work, a more affordable model is often enough here.
Pro Tip: A reasonable working configuration: set Planner and Layout to a strong, more expensive model, since the plan and the layout quality shape everything that follows; leave Agent on a solid mid-range model for carrying out the steps; and drop Ask down to the cheapest option, since it's only answering questions.
Matching the Model to the Job
A quick guide for picking a setting instead of guessing:
Task | Recommended setting |
A typo sweep across existing text | Auto Value |
An alt-text pass on a batch of images | Auto Value |
Restyling a section's design | Auto Balanced |
Building a full page from a screenshot | Auto Premium |
An architectural question about how your site is put together | Auto Deep Value |
If none of these quite fit your needs, start with Auto Balanced and move up or down a tier once you've seen how the Agent handles similar requests.
Pro tip: The model you pick is one of the biggest factors in how fast your balance goes down - a Premium-tier model on every request will draw down faster than Auto Value on the same work. See billing for how usage is calculated and what draws on your balance.