Verify Your Claude Key in OpenClaw
Set OpenClaw to use Anthropic API-key authentication, then send a test message in a fresh session and check the active model with /status. A saved model selection alone cannot tell you whether the live request used the intended Anthropic account.
Choose the authentication path first
Start with a key from the Claude Platform API Keys page for the workspace that should own the usage. Give the key a name that identifies the OpenClaw instance and store it in your secret manager. Open the model settings for the target instance, choose Anthropic, and select the API-key option. If the setup offers a gateway location, choose Local when the gateway runs on that same instance.
OpenClaw also documents a Claude-login path with Extra Usage enabled. If your goal is to use a specific API workspace, choose the API-key path explicitly and supply the key only through the instance's credential input. Before saving, check that you're editing the intended instance and that the setup identifies Anthropic as the provider.
Use different trial IDs for each authentication path. A cached reply or an old session can otherwise make a failed route appear healthy.
API usage and costs - OpenClawSelect a model the instance offers
Use the model picker on the target instance to choose the exact Anthropic model you want. Anthropic identifies Sonnet 5 and Sonnet 4.6 as distinct generations, so record the full model identifier offered by the picker. A family name such as Sonnet is too broad to use as the expected result of an acceptance test.
If your preferred model is absent, first confirm that the instance is configured for Anthropic. Then check model availability for the API workspace and model support for the installed OpenClaw release. If you choose an available Sonnet version to finish testing the key, record that choice and keep the preferred model pending. A successful call on the substitute verifies that path for the substitute model.
Record the model offered by the running instance and the billing route you intended. Reconcile the same synthetic request in the appropriate usage view.
Claude Sonnet \ AnthropicProve the path with one fresh turn
In this illustrative test, assume the intended instance is named Intake Test, the Claude Platform workspace is named Operations, and the model picker offers Anthropic Sonnet 5. Start a fresh OpenClaw session on Intake Test and send the plain-text request "Reply with CHECK-17 only." Use text with no attachments or external actions so the test covers a single model turn.
After the reply, run /status in that session. OpenClaw reports the session model and last-response token information there. The test passes for the model when the reply returns CHECK-17 and /status shows the exact selection you recorded. For account attribution, also check whether usage appears in Operations during the test window. If the workspace has concurrent traffic or usage has yet to appear, leave attribution pending.
API usage and costs - OpenClawDiagnose the first failed check
If a fresh session replies under an unexpected model, revisit the instance default and any session-level selection before touching credentials. If the intended model is selected and the provider rejects authentication, check that the key came from the intended workspace and was saved through the Anthropic API-key option. For a model availability error, confirm that the exact model is available to the workspace and supported by the installed build.
OpenClaw's per-message usage footer can include an estimated cost when local pricing and usage metadata are available. A missing estimate therefore gives you a separate display issue to investigate. Keep the test result tied to the reply and active model, and use the provider's records when you need billing attribution.
API usage and costs - OpenClaw