# Choosing a Model How to select the best AI model for a published Chipp chat agent. Model selection applies to conversational agents -- one output type Alchemist supports. For non-chat workloads (APIs, backend services, data pipelines), the AI runs in the engineering pipeline, not the deployed software. --- Chipp supports multiple AI models from OpenAI, Anthropic, Google, and more. Each has different strengths, speeds, and costs. This guide helps you choose the right one. ## Quick Recommendations | Use Case | Recommended Model | |----------|-------------------| | General purpose | GPT-5.4, Claude Sonnet 4.6, or GPT-5.6 Terra | | Image analysis | Gemini 3.1 Pro, GPT-5.4, or Claude Sonnet 4.6 | | Complex reasoning / frontier work | o3, Claude Opus 4.8, GPT-5.5 Pro, or GPT-5.6 Sol | | Fast responses | Gemini 3.5 Flash, GPT-5 Mini, or Claude Haiku 4.5 | | Long documents | Gemini 3.1 Pro or GPT-4.1 (1M token context) | | Cost-sensitive / high-volume | Gemini 3.5 Flash, GPT-5 Nano, or GPT-5.6 Luna | ## Estimate Your Costs Use our calculator to estimate monthly AI costs based on your expected usage: ## Featured Models ### Best for General Purpose These models excel at a wide range of tasks including writing, coding, analysis, and conversation. ### Best for Speed When response time matters most, these models deliver near-instant results without sacrificing quality. ### Best for Reasoning For complex problem-solving, multi-step analysis, and tasks requiring deep thinking. ### Best Value for Long Documents Process entire codebases, legal contracts, or book-length content with massive context windows. ## Model Deep Dives ### OpenAI GPT-5.4 ### OpenAI GPT-5.6 Sol GPT-5.6 Sol is OpenAI's newest frontier model (also selectable via the `gpt-5.6` alias). It targets the same complex-reasoning and coding use cases as GPT-5.5 Pro and Claude Opus 4.8. Its price and context window are in the table under [Compare prices](#compare-prices). See [Model Benchmarks](https://developers.openai.com/api/docs/models/gpt-5.6-sol) for official specs. ### OpenAI GPT-5.6 Terra GPT-5.6 Terra balances intelligence and cost: a general-purpose option priced below both GPT-5.4 and GPT-5.6 Sol. ### OpenAI GPT-5.6 Luna GPT-5.6 Luna is OpenAI's most cost-efficient GPT-5.6 tier, aimed at high-volume, cost-sensitive workloads. ### Claude Sonnet 4.6 ### Gemini 3.1 Pro ### OpenAI o3 ### Claude Haiku 4.5 ### Gemini 3.5 Flash ## Compare Prices Prices per 1 million tokens, including the platform markup, and each model's context window: | Model | Provider | Input per 1M tokens | Output per 1M tokens | |-------|----------|---------------------|----------------------| | GPT-5.4 | OpenAI | $3.25 | $19.50 | | GPT-5.6 Sol | OpenAI | $5.20 | $26.00 | | GPT-5.6 Terra | OpenAI | $2.60 | $15.60 | | GPT-5.6 Luna | OpenAI | $0.26 | $1.56 | | Claude Sonnet 4.6 | Anthropic | $3.90 | $19.50 | | Gemini 3.1 Pro Preview | Google | $2.60 | $15.60 | | o3 | OpenAI | $2.60 | $10.40 | | Claude Haiku 4.5 | Anthropic | $1.30 | $6.50 | | Gemini 3.5 Flash | Google | $1.95 | $11.70 | | Model | Provider | Context window | Max output | |-------|----------|----------------|------------| | GPT-5.4 | OpenAI | 128,000 tokens | 32,768 tokens | | GPT-5.6 Sol | OpenAI | 128,000 tokens | 128,000 tokens | | GPT-5.6 Terra | OpenAI | 128,000 tokens | 128,000 tokens | | GPT-5.6 Luna | OpenAI | 128,000 tokens | 128,000 tokens | | Claude Sonnet 4.6 | Anthropic | 1,000,000 tokens | 64,000 tokens | | Gemini 3.1 Pro Preview | Google | 1,048,576 tokens | 65,536 tokens | | o3 | OpenAI | 200,000 tokens | 65,536 tokens | | Claude Haiku 4.5 | Anthropic | 200,000 tokens | 64,000 tokens | | Gemini 3.5 Flash | Google | 1,048,576 tokens | 65,536 tokens | ## Cost-Effective Options If you're optimizing for cost, these models offer excellent value: ## Key Considerations ### Vision Support If your app analyzes images, choose a model with native vision support. Models without vision use a fallback that may be less accurate. **Vision-capable models:** - All GPT-4.1 and GPT-5.x variants, including the GPT-5.6 family (not o-series reasoning models) - All Claude models - All Gemini models **No vision support:** - OpenAI o-series (o1, o3, o4-mini, etc.) ### Response Speed Speed matters for user experience. Faster models keep conversations flowing naturally. **Fastest:** GPT-5 Nano, Claude Haiku 4.5, Gemini 3.5 Flash, GPT-5.6 Luna **Medium:** GPT-5.4, Claude Sonnet 4.6, GPT-5.6 Terra **Slower:** Claude Opus 4.8, o1, o3, GPT-5.6 Sol (reasoning takes time) ### Context Window For processing long documents, choose models with large context windows: **1M+ tokens:** GPT-4.1 variants, GPT-5.x variants (including GPT-5.6 Sol/Terra/Luna at 1.05M), Claude Opus 4.8, Claude Sonnet 4.6, all Gemini models **200k tokens:** Claude Haiku 4.5, OpenAI o-series Note: Chipp enforces a 128K runtime ceiling on the entire GPT-5.x family (including GPT-5.6) regardless of the published context window, to stay within OpenAI's Responses API's practical limits. ### Reasoning Quality For complex tasks requiring multi-step reasoning: **Best reasoning:** o3 Pro, Claude Opus 4.8, GPT-5.5 Pro, GPT-5.6 Sol **Very good:** GPT-5.4, Claude Sonnet 4.6, o3, o4-mini, GPT-5.6 Terra **Good:** GPT-5 Mini, Claude Haiku 4.5, Gemini 3.1 Pro, GPT-5.6 Luna ## HIPAA-Compliant Models Building a healthcare app? Chipp holds BAAs with Google (Gemini models) and Anthropic (Claude models, excluding Claude Fable). Choose one of these BAA-covered models for your app's chat. Enabling **HIPAA Compliance** in **Launch > Access** enforces authentication and session controls, but it does not itself restrict which chat model you can pick, so make sure you selected a covered model. See [HIPAA-Compliant Models](/docs/guides/hipaa-compliant-models) for the full breakdown and important caveats. ## Changing Your Model 1. Go to your app in the Chipp dashboard 2. Navigate to **Brain** > **Prompt** 3. Under **Chat Model**, select your preferred model 4. Your change saves automatically Changes take effect immediately for new conversations. ## Testing Different Models Not sure which model works best? Try these approaches: 1. **A/B testing**: Create two versions of your app with different models and compare user feedback 2. **Specific prompts**: Test your most common use cases with different models to see quality differences 3. **Speed vs. quality**: Start with a fast model, then upgrade if users need better responses ## Using with MCP You can manage AI models programmatically using the Chipp MCP Server, including AI-powered model recommendations. ### Available Tools | Tool | Description | |------|-------------| | `get_model` | Get current model for an app | | `update_model` | Change the AI model for an app | | `list_available_models` | List all available AI models | | `recommend_model` | Get AI-powered model recommendations | ### Example Prompts ``` "What model is my FAQ bot using?" "Switch my support bot to GPT-5.4" "Which model would you recommend for my customer service app?" "List all available models with their pricing" ``` ### Workflow: AI-Powered Model Selection ``` You: Recommend a model for my customer support bot that needs to be fast and handle simple FAQ questions AI: [Calls recommend_model with priority="speed"] Based on your use case, I recommend: 1. **Gemini 3.5 Flash** (Best choice) - Very fast responses (~1.5s) - Cost: $1.50/M input, $9.00/M output - Great for FAQ and simple queries 2. **GPT-5 Nano** (Budget option) - Fastest response time - Cost: $0.05/M input, $0.40/M output - Good for high-volume, simple tasks Want me to update your app to use Gemini 3.5 Flash? You: Yes, update it AI: [Calls update_model] Updated FAQ Bot to use Gemini 3.5 Flash. Changes take effect for new conversations. ``` > **Note:** MCP server access requires a Studio plan or higher. See the [MCP Setup Guide](/docs/guides/mcp/setup) to get started.