Skip to content
Return to ConnectContact support

Choose a model for accounting work

An AI model is the part of an AI app that reads your request, decides which steps to take, and writes the response. Your app may offer several models: some are faster or cheaper, while others are better at following a long procedure and checking details.

For accounting work, choose the strongest general-purpose or reasoning model your app and plan make available. This matters when the app must inspect records, compare companies, trace a balance, follow several checks, or prepare a change for your approval.

Look for labels such as Thinking, Reasoning, High, Pro, Opus, or Sonnet. Names and access vary by app, plan, workspace, and rollout, so the picker in your app is the current source of truth.

The strongest option may take longer or use more of your plan, but that tradeoff is usually worthwhile when the workflow touches accounting records.

Avoid the cheapest or fastest option as your default for multi-step work. It may be labeled Haiku, Mini, Lite, Instant, Fast, or Economy. A smaller model can miss a condition, stop too early, or lose track of the company or period.

Fast or lower-cost models are useful after the hard work is done: summarize findings, rewrite a note, or draft a follow-up question. Use a stronger model when the app is investigating records, deciding what to inspect next, or preparing a proposed accounting change.

No model is a substitute for review. Confirm the workspace, company, period, evidence, and proposed changes before approving anything that could change accounting data.

Some clients choose a model automatically, hide the picker behind a plan or administrator setting, or expose it only in a configuration file. If you cannot choose, use the strongest default the client provides and ask the app which model it is using when the answer matters. The setup page for your client explains any client-specific choice.