AI models
Pick the model that powers your agent. Each one trades cost against capability.
Open your agent's Sandbox or Publish page, then find the Model dropdown →
Choosing a model
Each agent runs on one model at a time. Open the Model dropdown in the Sandbox or on the Publish page, then pick from the catalogue. Models are grouped by tier, and every option shows its per-message credit cost so you can budget at a glance.
Your agent needs at least one trained knowledge source before you can switch models. Until then the dropdown stays disabled with a prompt to add knowledge first.
Free vs paid models
Free plans can use three models, all costing 1 credit per message:
- GPT-4.1 Nano
- GPT-4o Mini
- GPT-5.4 Nano
Every paid plan, from Starter upward, unlocks the full catalogue, including frontier models from OpenAI, Anthropic, and Google. Paid models appear in the free dropdown with a lock icon, clicking one opens an upgrade prompt rather than selecting it.
Credit cost per message
Each reply your agent sends consumes credits. The most efficient models cost 1 credit per message, the most powerful cost up to 25. The selector shows the cost next to every option, written as "1/msg", "10/msg", and so on.
| Tier | Example models | Credits per reply |
|---|---|---|
| Fast and cost-efficient | GPT-4.1 Nano, GPT-4o Mini, GPT-5.4 Nano, GPT-5.6 Luna, Gemini 3.1 Flash-Lite, Gemini 3 Flash, Claude Haiku 4.5 | 1 to 5 |
| Versatile and highly intelligent | GPT-5.6 Terra, GPT-5.4, GPT-5.2, GPT-5.1, GPT-4o, Gemini 3.1 Pro, Claude Sonnet 4.6 | 10 to 15 |
| Most powerful at complex tasks | GPT-5.6 Sol, Claude Opus 4.7, Claude Opus 4.6, GPT-5.5 | 25 |
The GPT-5.6 family
GPT-5.6 from OpenAI is available on every paid plan, from Starter upward. It arrives as three tiers rather than one model, so you can match each agent to the right balance of capability, speed, and cost. Pick any of them from the Model dropdown, and switching models never changes what your agent has already learned.
- GPT-5.6 Sol is the flagship, the most capable of the three for complex or high-stakes questions. It costs 25 credits per message.
- GPT-5.6 Terra is the balanced default for most agents. OpenAI positions it as competitive with the previous GPT-5.5 flagship while costing less to run, so it delivers premium-class quality at 10 credits per message.
- GPT-5.6 Luna is the fastest, built for high-volume, everyday questions where responsiveness matters most. It costs 5 credits per message.
A simple way to choose: start on Terra, then move up to Sol for detailed or sensitive questions, or down to Luna for high-traffic, routine ones. Because the model is set per agent, you can run a different tier on each agent without retraining anything.
Unlocked models surface first
Within each tier the dropdown lists the models you can pick above the locked ones, so a free user sees their available choices without scrolling past upgrade rows.
Compare two models
Below the Model dropdown in the Sandbox, "Compare two models on the same question" runs the same prompt through two models side by side so you can judge quality before you commit. Each side spends credits at its own model's rate.
Troubleshooting
The model dropdown is greyed out
Symptom. The Model dropdown won't open, or it shows an "add knowledge first" hint instead of the catalogue.
Cause. A brand-new agent has nothing to answer from yet, so the model picker stays disabled until the agent has at least one trained source to reason over.
Fix. Add a knowledge source and let it finish training (a website URL, a file, text, or a Q&A pair). Once the first source is trained the dropdown unlocks. See Knowledge sources.
Credits are draining faster than expected
Symptom. Your balance drops by several credits per reply, far quicker than the "1 credit is roughly one message" rule of thumb.
Cause. The agent is set to a premium model. Anything above the 1-credit tier costs more per reply: a 2-credit model doubles your burn, and a top-tier model spends up to 25 credits on a single answer.
Fix. Open the Model dropdown and check the "/msg" cost next to the selected model. Switch to a 1-credit model (GPT-4o Mini, GPT-4.1 Nano, or GPT-5.4 Nano) for everyday traffic, and reserve the premium models for agents that genuinely need the extra reasoning. See Understanding credits.
Still stuck? We'll help.
Every plan, including Free, comes with human support. If something isn't behaving the way you expect after following the steps above, send us a note and we'll get back to you within one business day.