Enable Anthropic Claude for AI grading
QuizMachine uses Anthropic's native Messages API to suggest grades for free-text and code answers, and to review question wording. Every score remains subject to administrator review.
Your API key is encrypted at rest and never displayed after saving. A fallback may send the same answer to another selected provider and each attempt can be billable.
Step 1 — Create an API key
- Sign in to the Claude Console API keys page.
- Create a workspace-scoped key for QuizMachine, choose an appropriate expiration, and copy it immediately.
- Confirm billing, spend limits, and access to the model you intend to select.
Step 2 — Choose endpoint and model
| Setting | Value |
|---|---|
| Endpoint | https://api.anthropic.com/v1/messages |
| Default model | claude-haiku-4-5 |
Choose another Messages API model only after checking Anthropic's current model list. Claude 3 Sonnet, Opus, and Haiku are legacy model choices only, where your account still exposes them; they are not separate providers.
This initial integration supports free-text and code grading plus question review. Diagram Upload questions go to manual review rather than receiving a text-only grade — but only once no configured provider has graded them: if you have enabled further providers, a declined diagram is first offered to the next provider in priority order.
Step 3 — Add it in QuizMachine
- Open the quiz Configure tab or your personal AI settings.
- Choose Anthropic (Claude) for a fallback priority.
- Enter the model, optional endpoint, and API key.
- Use Test configuration, then save the fallback profiles.
Troubleshooting and rotation
- HTTP 401 — the key is invalid, expired, revoked, or lacks workspace access.
- HTTP 429 — the workspace reached a rate or spend limit.
- HTTP 400 — check the model name and custom endpoint.
- Manual review — no AI score was applied after a safe request or parsing failure.
Create and test a replacement key first, update the QuizMachine profile, then revoke the old key in the Claude Console. Disable or remove the profile when it must no longer receive answer data.
Reference: Anthropic Messages API and authentication guidance.