The best model is not automatically the strongest. Sol fits complex briefs and difficult decisions, Terra balances quality and cost for daily work, and Luna suits fast or repetitive processing.
The useful answer depends on the purpose of the page, the people allowed to edit it and the cost of an error. A reliable workflow separates editorial decisions, technical delivery and final verification instead of hiding all three inside one automated operation.
The decision to make first
Classify each stage by complexity, volume and risk. One article can use Sol for planning and verification, Terra for drafting and Luna for simple metadata.
A practical method
- Define audience, angle, length, sources and rejection criteria.
- Build three representative tests before processing the batch.
- Compare accuracy, tone consistency, review time and total cost.
- Separate generation, factual checking and editorial approval.
- Record the model and instructions used for every item.
Checks before publishing
- The source of truth is explicit and can be restored.
- URLs, dates and visible claims match the delivered page.
- A human reviewer can reproduce the check without a hidden service.
The main pitfall
Changing models without replaying tests can alter length, style and caution. Record the evaluated version for critical workflows.
Recommended next step
Create a task × model × review-level matrix and test it on five real pieces before scaling.
Compare models on an editorial task
A team gives three models the same source pack and instructions. It asks for a service-page summary with attributed facts and unresolved points. Reviewers read the outputs without model names and record factual errors, omissions, clarity and correction time.
Choose on task performance and the full cost of review, not a model name or a single successful demo. Keep the sources and prompt version so the comparison can be repeated as models change.