# AI Model Selection Choosing the right model for a given task. The model landscape is broad and the right choice depends on the use case, not the hype cycle. ## Selection criteria - **Capability**: can the model actually do the task? Coding, reasoning, creative writing, and classification require different strengths. - **Cost**: [[AI Frontier Model|frontier models]] can be 100x more expensive than [[Small Language Models (SLMs)]]. See [[AI Cost Management]]. - **Latency**: real-time applications need fast inference. Smaller models and local deployment win here. - **Privacy**: sensitive data may require [[Running AI Models Locally|local models]] or private deployments. - **[[Context Window]]**: long documents need models with large context windows. - **Tool support**: agentic workflows need models with strong function calling capabilities. ## The model landscape - **[[AI Frontier Model|Frontier models]]**: best capabilities, highest cost, API-only. - **[[AI Open Weight Models|Open weight models]]**: strong and improving fast, can self-host, full control. - **[[Small Language Models (SLMs)]]**: fast, cheap, good enough for many tasks. Often the right answer. - **[[Running AI Models Locally|Local models]]**: maximum privacy and zero marginal cost after setup. ## Matching model to use case Simple classification, extraction, and formatting tasks do not need frontier models. [[Model routing]] enables dynamic selection: route each request to the cheapest model that can handle it. Some tools now make that choice for you: [[GitHub Copilot Fusion]] picks the model (and sometimes several) per prompt. [[AI Evaluation]] before committing to a model avoids expensive mistakes. ## When a model falls short: escalate in order Anthropic's developer team (ClaudeDevs, November 2026) gives an order to follow when a model doesn't do the job well enough, before paying for a bigger one: 1. Give the model a way to verify its own work (tests, a checker, screenshots) 2. Raise effort by one level 3. Move up a model, starting at a lower effort The reasoning: past `high`, one more effort level can cost as much as the bigger model, and "moving to a higher effort level won't give a model knowledge or skill it doesn't have." That's why they say "a higher price per token doesn't always mean a higher cost per task." Their example: before raising [[Claude Opus 5.5]] past `high` on a hard task, test [[Claude Fable 5.2]]. The same guide keeps subagents on smaller models (Haiku) for mechanical work. ## References - ClaudeDevs, "Which Claude model should you build with?": https://claude.dev/blog/which-claude-model-should-you-build-with ## Related - [[AI Frontier Model]] - [[AI Open Weight Models]] - [[Small Language Models (SLMs)]] - [[Running AI Models Locally]] - [[Model routing]] - [[GitHub Copilot Fusion]] - [[AI Cost Management]] - [[AI Evaluation]] - [[Context Window]] - [[Claude Fable 5.2]] - [[Claude Opus 5.5]]