Models
Choose models for your tasks using document drafting and editing performance and tool call records.
Choose a model by checking both the result and usage. Compare the same request and source material to identify omissions and editing mistakes.
Start with a small task
Claude Haiku 5.5 and GPT-6 Luna are available on the Free plan and can be used for a short draft or a specific edit. For more complex material, compare the results with another available model.
- Name the document, the section to change, and what to preserve.
- Check facts, citations, and the saved result.
- If the result is insufficient, compare the same request with another model or reasoning effort.
Scope of measurements
Compare measurements from the same model version and task conditions. A controlled comparison of every current model needs further measurement. GPT-6.1 Terra is unmeasured in this guide. Treat earlier GPT-5.6 Terra measurements as evidence for that version, and compare current models using speed, accuracy, and cost from your actual tasks.
Choose reasoning effort
Start with a supported lower level and increase it when comparing sources or developing a complex explanation. The Free plan uses Off. Claude Opus 5.5 and GPT-6.1 Sol start at Low.


Check consistency between code and documents
Documents and canvases linked to repositories can be updated in place after a push. After an automatic update, compare changed defaults, removed features, paths, and citations with the current code. Check that unrelated manually written content remains intact.
See Model overview, Usage and AI budget, and Document automation.