Choose a model

The model is the reasoning engine behind an agent's responses. It uses the same system prompt and sources as other models, but models differ in instruction following, complex reasoning, latency, and credit cost. Start with the default model. Test it and record a baseline. If your own conversations show a real improvement, switch models.

Where to choose a model

You can select a model in two places:

• Playground — test a model with a candidate system prompt before you save it.
• Settings → General — saves the agent's primary model after you select Save Changes.

In Playground, select Save to Agent to save the current model and system prompt together. If you only choose a model, or only select Apply & Test, you change the test configuration. The saved agent stays the same.

AgentStack currently uses GPT-5.2 as the default model. The Playground model selector groups models by provider and labels recommended options. It also shows the credit cost beside each option. The selector in Settings → General shows model names in a single list. Use the Playground selector as the current list of models: providers and model versions can change faster than this guide.

The model you save becomes the primary model. If its provider fails before it sends any response content, AgentStack can run a configured fallback model. Billing still uses the fixed credit rate of the primary model.

How to choose

Use these criteria in order:

1. Correctness on your acceptance set. The model must retrieve and apply your policies accurately.
2. Boundary behavior. The model must follow the agent's instructions for uncertainty and escalation.
3. Consistency. Repeat important prompts. Look for unstable answers.
4. Latency. Faster responses matter for high-volume, conversational support.
5. Credits. Choose the least expensive model that reliably meets the first four criteria.

Smaller one-credit models can be a strong fit for direct, well-grounded questions. When requests need complex reasoning or careful instruction following, test higher-credit models. Credit cost does not show a model's context capacity. A higher price is not proof of better performance on your workload.

Run a controlled comparison

Open Playground. Select Compare Models. Comparison mode sends one shared message to every active panel. Each panel has its own model, temperature, and system prompt. The interface shows the combined credit cost of the active panels.

To compare models fairly:

1. Use the same system prompt in every panel.
2. Set the same temperature.
3. Send the same acceptance prompts.
4. Grade facts, boundary behavior, and completeness with a simple pass/fail rubric.
5. Repeat the hardest prompts.
6. If the winning panel improves the full set, save it to the agent.

Comparison mode supports two to four panels. You can pause a panel to exclude it from shared sends and the displayed cost. The browser keeps panel configurations locally for the agent. The agent's live configuration changes only after you save a panel to it.

Understand credits

The credit number beside a model is the charge for one message. The selector uses this number to calculate cost. A comparison message costs the sum of the model credits for all active panels. Playground shows this total cost before you send the message. For example, when you compare two active models, one test message runs both models and charges both credit costs.

Credits make repeated tests measurable. Do not skip evaluation to spend fewer test messages. A short acceptance run costs less than an agent that gives customers wrong answers. See Billing and credits for balances, plan allocations, and top-ups.

When a model change will not help

A model change rarely fixes a missing, stale, or contradictory source. It also cannot make an ambiguous system prompt precise. Follow the response-quality workflow first. Verify the knowledge. Remove conflicts. Clarify agent behavior. Then compare models on the remaining hard cases.