LLM Selection Framework

A decision framework for choosing which LLM to use for a given task, built on cost-performance analysis across GPT-4o, Claude 3.5/3.7, Gemini 1.5 Pro, and Llama 3 variants.

The framework has two parts: a task classification system that maps task types to model strengths across seven categories, and a cost calculator that estimates spend per task type at different usage volumes.

Key findings: Claude 3.5 Haiku is underused for high-volume extraction tasks where it outperforms larger models at a fraction of the cost. Smart routing vs single-model defaults typically produces a 40-60% cost reduction at equivalent output quality.

What is included: the framework document, the task classification table, a cost comparison table (updated Q1 2026), and a worked example routing a 10-task workflow through the framework.

Best for operators paying API bills who want to route tasks intelligently rather than defaulting to one model for everything.