Model Selection Guide
Choose a model by task requirements, data path, provider limits, and your own results.
Overview & Objectives
Compare browser-local runtimes with connected model providers. Choose by the task you need to complete, the information the model can accept, its provider limits and costs, and how your own representative prompts perform.
Core Tutorial Steps
Define the job first
Write down the task, the expected result, and how you will check it. A model that is strong on a public benchmark may not be the best fit for your own workflow.
Choose a local or connected model
A browser-local model processes its prompt in the browser after model files are available. A connected provider receives the request under that service’s terms; some app features route requests through an Accelerated Logic endpoint.
Check supported input and context
Read the provider’s current model documentation for supported text, image, audio, or video input and its context limit. Do not assume that a large context limit guarantees accurate retrieval.
Compare cost, latency, and rate limits
Check current provider pricing, rate limits, caching rules, and output charges. Include retries and human review when you estimate the cost of a task.
Run the same representative prompts
Try a small set of real tasks, record what succeeded and failed, and compare outputs against your stated quality bar. Keep vendor benchmark results separate from your own observations.
Recheck the provider’s official documentation when a model version, price, or data-use term changes.