> ## Documentation Index
> Fetch the complete documentation index at: https://docs.a2v2.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Model & Temperature

> Choose the AI model that powers your agent and tune how creative its answers are.

Your agent's **model** determines the quality, speed, and credit cost of its
answers. **Temperature** controls how creative versus focused those answers are.
Both are set on the [Sandbox](/testing/sandbox).

## Prerequisites

* An agent (see [Create an agent](/building/create-agent)).

## Choose a model

<Steps>
  <Step title="Open the Sandbox">
    In the agent sidebar, open **Agent → Sandbox**.
  </Step>

  <Step title="Open the model selector">
    Under **Choose Your AI Model**, open the selector. Each option shows its
    **credits per message** and any badges (**HIPAA**, **Premium**).
  </Step>

  <Step title="Pick a model">
    Choose the model that fits your needs. New agents start on **Claude Haiku
    4.5** (1 credit/msg).
  </Step>

  <Step title="Save changes">
    Select **Save Changes** to apply your selection to the agent's live answers.
  </Step>
</Steps>

<Frame>
  <img src="https://mintcdn.com/a2v2ai/ItYQQYTHsiiRmr-t/images/building/model-selector.png?fit=max&auto=format&n=ItYQQYTHsiiRmr-t&q=85&s=c9c7ade4c0db3139556d0dc1c86b7f6f" alt="The Choose Your AI Model selector showing models with credit costs and HIPAA/Premium badges." width="1440" height="900" data-path="images/building/model-selector.png" />
</Frame>

<Note>
  Different models consume different amounts of **credits per message**. A more
  capable model gives better answers but costs more per reply — see
  [Credits](/concepts#credits).
</Note>

### Choosing the right model

| If you want…                     | Lean toward                               |
| -------------------------------- | ----------------------------------------- |
| The best answer quality          | A more capable (higher-cost) model        |
| Faster, cheaper replies at scale | A lighter, lower-cost model               |
| A balance                        | Start mid-tier, then adjust after testing |

<Tip>
  Use the [Playground](/testing/playground) to run the same questions against two
  models side-by-side before committing.
</Tip>

## Model reference

The models below are available in the **Choose Your AI Model** selector. The
**HIPAA** badge marks models eligible for HIPAA use; the **Premium** badge marks
higher-tier models. Credit cost is charged per message.

| Model                          | Badges         | Credits/msg |
| ------------------------------ | -------------- | ----------- |
| GPT-5.4                        | Premium        | 2           |
| GPT-5.4-mini                   | —              | 1           |
| GPT-5.4-nano                   | —              | 1           |
| GPT-5.1                        | Premium        | 1           |
| GPT-5                          | Premium        | 1           |
| GPT-5-mini                     | —              | 1           |
| GPT-5-nano                     | —              | 1           |
| Llama 3.3 70B Versatile        | HIPAA, Premium | 1           |
| Llama 3.1 8B Instant           | HIPAA          | 1           |
| Qwen 3 32B                     | HIPAA, Premium | 1           |
| GPT OSS 120b                   | HIPAA, Premium | 1           |
| GPT OSS 20b                    | HIPAA, Premium | 1           |
| Gemini 2.5 Pro                 | HIPAA, Premium | 2           |
| Gemini 2.5 Flash               | HIPAA          | 1           |
| Claude Opus 4.6                | HIPAA, Premium | 3           |
| Claude Sonnet 4.6              | HIPAA, Premium | 2           |
| **Claude Haiku 4.5** (default) | HIPAA          | 1           |
| GLM-5                          | HIPAA, Premium | 1           |

<Note>
  The model catalog and credit costs can change over time — the selector in the app
  is always the source of truth for what's available and what it costs.
</Note>

## Tune temperature

**Temperature** sets how much the model varies its wording:

| Temperature | Behavior                        | Good for                             |
| ----------- | ------------------------------- | ------------------------------------ |
| **Low**     | Focused, consistent, repeatable | Support, policies, factual answers   |
| **Higher**  | More varied and creative        | Brainstorming, casual/marketing tone |

Adjust it with the **Temperature** slider on the Sandbox.

<Warning>
  For support and factual agents, keep temperature **low**. High temperature makes
  answers less predictable and more likely to drift from your sources.
</Warning>

## Tips

<Tip>
  * Pair a capable model with **low temperature** for accurate, grounded support
    answers.
  * If answers feel robotic, nudge temperature up slightly — but re-test for
    accuracy.
  * Model choice and temperature interact with your [instructions](/building/instructions);
    tune them together.
</Tip>

## Related

<CardGroup cols={2}>
  <Card title="Write instructions" icon="pen" href="/building/instructions" />

  <Card title="Compare in the Playground" icon="code-compare" href="/testing/playground" />

  <Card title="How credits work" icon="coins" href="/concepts#credits" />

  <Card title="Skills" icon="wand-magic-sparkles" href="/building/skills" />
</CardGroup>