Run the same prompt across multiple models and temperatures. Results are cached automatically — re-runs are instant.