> ## Documentation Index > Fetch the complete documentation index at: https://docs.arcbeam.ai/llms.txt > Use this file to discover all available pages before exploring further. # Model Selection > Choosing the right AI model for your needs **Model selection** is about choosing which AI "brain" to use for different tasks. Just like you wouldn't use a calculator for writing an essay, different AI models are better suited for different jobs. Each AI model has trade-offs in terms of quality, speed, and cost. The key is finding the right balance for your use case. ## Understanding AI Models Smart assistants who can handle routine questions Subject matter experts who can tackle complex problems Specialists trained for specific tasks ## Popular AI Model Providers The major AI model providers offer a range of models with different capabilities and price points: The company behind ChatGPT offers models ranging from the premium GPT-4o to the cost-effective GPT-4o mini. Known for strong general-purpose performance. **Visit:** [OpenAI Platform](https://platform.openai.com/docs/models) Known for following instructions well and handling longer conversations. Offers models from Claude Sonnet 4.5 (premium) to Claude Haiku 4.5 (fast & affordable). **Visit:** [Anthropic Models](https://docs.anthropic.com/en/docs/models-overview) Can handle extremely long conversations and documents with large context windows. Latest models include Gemini 3 Flash and Gemini 2.0 Pro. **Visit:** [Gemini API](https://ai.google.dev/gemini-api/docs/models) Models you can run on your own servers offer complete data privacy and control, but require technical setup and infrastructure management. **Popular Models:** * **Llama** - Meta's open-source models * **Mixtral** - Mistral AI's mixture-of-experts models * **Qwen** - Alibaba's Qwen series **Benefits:** Complete data privacy, no per-request costs, full customization **Tradeoffs:** Requires infrastructure, technical expertise, and maintenance **Visit:** [Hugging Face Models](https://huggingface.co/models) **Model selection changes frequently.** Visit the provider documentation above for the latest models, pricing, and capabilities. Use [evaluations](/v0/core-concepts/evaluations) to test which models work best for your specific use case. ## How to Choose the Right Model ### Consider What You Need | Use Case | What to Choose | Example | Recommended Models | | ----------------------------- | -------------------------------------------------- | -------------------------------------------------------------------- | ---------------------------------------------------------- | | **Simple, High-Volume Tasks** | Cheaper, faster models | Answering basic FAQs, categorizing requests | GPT-4o mini, Claude Haiku 4.5, or Gemini 2.0 Flash-Lite | | **Complex Reasoning** | Premium models | Analyzing contracts, solving complex problems | GPT-4o, Claude Sonnet 4.5, or Gemini 3 Flash | | **Very Long Documents** | Models with large "memory" | Summarizing 100-page reports | Gemini 2.0 Pro (2M tokens) or Gemini 2.0 Flash (1M tokens) | | **Budget-Conscious Projects** | Most cost-effective model that meets quality needs | Start with GPT-4o mini and test if it's good enough before upgrading | Gemini 3 Flash ($0.50/$3 per 1M tokens) or GPT-4o mini | ### The Three-Factor Balance The key: Find the cheapest model that meets your quality and speed requirements. Every model choice involves balancing three factors: * How good are the responses? * How often is it correct? * Does it understand nuance? * How fast does it respond? * Can users wait that long? * Does it meet your performance needs? * How much does each response cost? * How many responses do you need per day? * Does it fit your budget? ## Common Model Selection Strategies ### Start with Mid-Tier Begin with a mid-tier model like GPT-4o mini that offers good quality at reasonable cost Run actual use case questions through the model to see how it performs Only move to a premium model if quality isn't meeting your requirements Only move to a cheaper model if costs are too high and quality allows **Why this works**: 80% of tasks work fine with mid-tier models ### Use Different Models for Different Tasks You don't need to use the same model for everything: **Example for a customer service AI**: **Use**: Cheaper model **Reason**: Saves money on high volume simple tasks **Use**: Premium model **Reason**: Quality matters more for sensitive issues **Use**: Mid-tier model **Reason**: Balance of quality and cost for moderate complexity ### Try Before You Commit Test with 50-100 example questions that represent your real use case Compare responses from different models side by side Evaluate quality, speed, and estimated cost for each model Make your decision based on actual test data, not assumptions ## Evaluating Model Performance The best way to choose a model is to test it with real questions from your use case. Create 20-50 test questions that represent what users will actually ask, then compare how different models perform on accuracy, speed, and cost. **Quick evaluation checklist**: * Are answers accurate and complete? * Is the tone appropriate for your use case? * How fast does each model respond? * What would daily costs be at your expected volume? For a complete guide on evaluating models systematically, including setting up automated testing and measuring performance over time, see [Evaluations](/v0/core-concepts/evaluations). ## Managing Models Over Time ### Track Performance Monitor how your chosen model performs: **Weekly checks**: Track user satisfaction scores and feedback Monitor how often the model produces errors Ensure response times stay within acceptable range Track actual spending against budget ## Cost Considerations ### Understanding Pricing AI models typically charge per "token" (roughly 3/4 of a word): **What affects your costs**: How much context you provide with each request How long the AI's responses are How many requests you make per day Premium models cost more than standard models **Example calculation**: If you send 1,000 requests per day: * Average input: 500 words = \~650 tokens * Average output: 100 words = \~130 tokens * Using GPT-4o mini: \~\$1.50/day * Using GPT-4o: \~\$25/day ### Ways to Reduce Costs Don't send unnecessary context - summarize long history instead of including everything If you only need a short answer, specify that - don't let the model ramble Don't use premium models for simple tasks that cheaper models can handle For common questions, save and reuse answers to reduce duplicate processing ## Common Mistakes to Avoid **The mistake**: "We'll just use the best model for everything to ensure quality" **Why it's wrong**: Most tasks don't need the absolute best model. You'll spend 10x more for 5% better quality **Better approach**: Test if cheaper models work first. Only upgrade where quality truly matters **The mistake**: "This new model is supposed to be better, let's switch immediately" **Why it's wrong**: "Better" in general doesn't mean better for your specific use case **Better approach**: Always test with your actual questions before switching **The mistake**: Focusing only on quality and cost **Why it's wrong**: If users have to wait 10 seconds for a response, they'll leave **Better approach**: Define acceptable wait times upfront and only consider models that meet them **The mistake**: Choose a model once and forget about it **Why it's wrong**: Models, costs, and your needs all change over time **Better approach**: Review model performance monthly and be ready to optimize ## Getting Started ### Your First Model Selection * What tasks will your AI handle? * How many requests do you expect per day? * What's your quality threshold? * What's your budget? * How fast do responses need to be? * Gather 30-50 real example questions * Define what "good" answers look like * Include mix of easy and hard questions * Try 2-3 candidate models * Run your test questions through each * Measure quality, speed, and cost * Pick the best fit for your needs * Start with your chosen model * Track real-world performance * Collect user feedback * Adjust if needed ### Questions to Ask Your Team **Before choosing**: "How many requests will we process per day/month?" "What's our budget for AI costs?" "How quickly do responses need to be?" "What happens if the quality isn't perfect?" "Do we need features like image understanding?" **After launching**: "What's our actual cost so far?" "Are users happy with response quality?" "How does this compare to our estimates?" "Should we test other models to optimize?" ## Next Steps Learn how to manage conversation history Strategies to reduce AI costs Test and measure model performance Monitor model performance across your system