# Choosing a Model

How to pick the right AI model for your AVCodex agent.

---

AVCodex supports multiple AI models from OpenAI, Anthropic, Google, and more. Each has different strengths, speeds, and costs. This guide helps you pick the right one for the job, whether that is a Crestron programming helper, a field tech triage agent, or an RFP assistant pulling spec from manufacturer PDFs.

## Quick recommendations

| Use case | Recommended model |
|---|---|
| General purpose (programmer help, internal SOP agent) | GPT-5.4 or Claude Sonnet 4.6 |
| Image analysis (rack photos, error code screen captures, signal flow drawings) | Gemini 2.5 Pro, GPT-5.4, or Claude Sonnet 4.6 |
| Complex reasoning (RFP responses, design engineering math, control system logic) | o3, Claude Opus 4.6, or GPT-5.4 Pro |
| Fast responses (QR-code room support, quick lookups) | Gemini 2.5 Flash, GPT-5.4 Mini, or Claude Haiku 4.5 |
| Long documents (manufacturer manuals, AS-built packets, multi-room programming guides) | Gemini 2.5 Pro or GPT-4.1 (1M token context) |
| Cost-sensitive (high volume, simple FAQ-style agents) | Gemini 2.5 Flash Lite, Gemini 2.5 Flash, or GPT-5.4 Nano |

## Estimate your costs

Use the calculator in the dashboard to estimate monthly AI costs based on your expected usage volume.

## Featured models

### Best for general purpose

These models handle a wide spread of tasks: SOP writing, code assistance, conversational support, light analysis.

### Best for speed

When response time matters most (think a tech in a rack closet on a phone), these models return near-instant results without giving up much quality.

### Best for reasoning

For multi-step analysis, design problems, and tasks that need careful thinking through specs and constraints.

### Best value for long documents

Read entire programming guides, manufacturer manuals, or full RFP documents in one shot using massive context windows.

## Model deep dives

### OpenAI GPT-5.4

### Claude Sonnet 4.6

### Gemini 2.5 Pro

### OpenAI o3

### Claude Haiku 4.5

### Gemini 2.5 Flash

## Cost-effective options

If you are optimizing for cost (for example, a public-facing room support agent that fields hundreds of touch panel questions a day), these models give strong value.

## Key considerations

### Vision support

If your agent analyzes images (a tech sending a phone photo of a Crestron error screen, a project manager uploading a rack drawing), pick a model with native vision support. Models without vision fall back to a workaround that can be less accurate.

**Vision-capable models:**

- All GPT-4.1 and GPT-5.x variants (not the o-series reasoning models).
- All Claude models.
- All Gemini models.

**No vision support:**

- OpenAI o-series (o1, o3, o4-mini, etc.).

### Response speed

Speed shapes the user experience. A field tech standing in a closet wants the answer in two seconds, not twenty.

- **Fastest**: GPT-5.4 Nano, Claude Haiku 4.5, Gemini Flash Lite.
- **Medium**: GPT-5.4, Claude Sonnet 4.6, Gemini 2.5 Flash.
- **Slower**: Claude Opus 4.6, o1, o3 (reasoning takes time).

### Context window

For processing long documents (full manufacturer manuals, multi-room AS-built packets), pick a model with a large context window.

- **1M tokens**: GPT-4.1 variants, GPT-5.x variants, Claude Opus 4.6, Claude Sonnet 4.6, all Gemini models.
- **200k tokens**: Claude Haiku 4.5, OpenAI o-series.

### Reasoning quality

For complex tasks (RFP scoring, design engineering, multi-system signal flow analysis):

- **Best reasoning**: o3 Pro, Claude Opus 4.6, GPT-5.4 Pro.
- **Very good**: GPT-5.4, Claude Sonnet 4.6, o3, o4-mini.
- **Good**: GPT-5.4 Mini, Claude Haiku 4.5, Gemini 2.5 Pro.

## Changing your model

1. Open your agent in the AVCodex dashboard.
2. Go to **Build** > **Configure**.
3. Under **Model**, pick your preferred model.
4. Click **Save**.

Changes take effect immediately for new conversations.

## Testing different models

Not sure which works best for your use case? A few approaches:

1. **A/B testing**: Spin up two versions of your agent (one Sonnet, one GPT-5.4 for example) and compare feedback from your team.
2. **Specific prompts**: Test your most common queries (a sample Crestron SIMPL question, a Q-SYS Lua snippet, a Biamp Tesira config question) against multiple models to compare quality.
3. **Speed vs quality**: Start with a fast model. If your users need deeper answers, step up.

## Using with MCP

You can manage AI models through code using the AVCodex MCP Server, including AI-powered model recommendations.

### Available tools

| Tool | Description |
|---|---|
| `get_model` | Get the current model for an agent. |
| `update_model` | Change the AI model for an agent. |
| `list_available_models` | List all available AI models. |
| `recommend_model` | Get an AI-powered model recommendation. |

### Example prompts

```
"What model is my Crestron programming agent using?"
"Switch my field tech support agent to GPT-5.4"
"Which model would you recommend for a high-volume room support agent?"
"List all available models with their pricing"
```

### Workflow: AI-powered model selection

```
You: Recommend a model for my field tech support agent that needs to be fast
and handle simple Crestron error code lookups.

AI: [Calls recommend_model with priority="speed"]

Based on your use case, I recommend:

1. **Gemini 2.5 Flash** (Best choice)
   - Very fast responses (~1.5s)
   - Cost: $0.10/M input, $0.39/M output
   - Great for short lookups and FAQ-style traffic

2. **GPT-5.4 Nano** (Budget option)
   - Fastest response time
   - Cost: $0.26/M input, $1.63/M output
   - Good for high-volume, simple tasks

Want me to update your agent to use Gemini 2.5 Flash?

You: Yes, update it.

AI: [Calls update_model]
Updated Field Tech Support to use Gemini 2.5 Flash. Changes take effect
for new conversations.
```

> **Note:** MCP server access requires a Builder plan or higher. See the [MCP Setup Guide](/docs/guides/overview) to get started.

*AVCodex · Your AV expertise. Amplified by AI.*