Start with a question. Follow the evidence.

A practical path through model research, comparison, context evidence and token-cost planning.

Public research works without an account. Saving, model execution and account registration are not connected in the current public release.

1. Define the task

Write down the intended workflow, expected output, review process and unacceptable failures before selecting a model. A general intelligence ranking is not a healthcare or legal evaluation.

2. Inspect the configuration

Open the model catalog, inspect the full brief and follow its context source. Keep reasoning effort and hosting variants distinct. Unverified context, missing measurements and stale data stay visibly labeled.

3. Compare and budget

Choose up to four configurations for side-by-side comparison. Estimate input and output tokens separately. The token counter is approximate; pricing calculations exclude caching, tools, hosting and other charges.

4. Evaluate before deployment

Test a representative, non-sensitive set of tasks with documented acceptance criteria. Inspect errors, latency, cost, access controls and operational recovery. The website does not execute these tests or approve a deployment.