A Claude model in the AIFast Link catalog only means it may be visible to the current account; it does not prove that the native Claude Code protocol is available. Claude Code and OpenAI-compatible clients use different request protocols. Confirm the endpoint before spending API key quota or subscription allowance.
First determine whether this is a native Claude Code endpoint
| Current documentation or test evidence | What the evidence supports |
|---|---|
| Only GPT/Claude model names are listed | This confirms a model name in the catalog, not Claude Code compatibility |
Only https://www.aifast.link/v1 or Chat Completions is shown |
Test it with an OpenAI-compatible client; this does not establish a native Messages URL |
| Current Link docs explicitly list an Anthropic Messages endpoint | Configure it according to the docs, then test a minimal request and streaming |
POST /v1/messages returns success |
This proves only that one request succeeded; also inspect events, tool calls, errors and usage fields |
Do not apply .hk、.club or another gateway's endpoints and model permissions to .link. Check each domain, account and plan separately; an old screenshot does not prove the current Link account has the same configuration.
Four pre-integration checks
- Link's current public description lists the Anthropic Messages endpoint
/v1/messages; first check whether the selected model and key group allow this protocol. A declared endpoint does not mean your account has passed request tests. - Confirm that the model ID required by Claude Code is available to the current account. Display names, console IDs and API request IDs may differ; use the live catalog or API response.
- For Link,
ANTHROPIC_BASE_URLuses the site roothttps://www.aifast.link, without/v1. Check that the final Messages request path is/v1/messages, with no duplicated version prefix or redirect to a web login. - Start with a short request using a separate, low-quota test key. Record the time, redacted status, final request path and request ID. Never put the key in shell history, public issues or screenshots.
Verify each capability separately
Do not treat one successful chat as proof that the full Claude Code workflow works. Record at least these results separately:
| Check items | Evidence of success | First checks on failure |
|---|---|---|
| Minimal message | Valid text is returned and the model ID matches the account catalog | Authentication, model access and request path |
| Streaming response | The client receives protocol events continuously and terminates normally | SSE event format, buffering and timeouts |
| Tool Calls | Tool request and result-return structures are both parsed correctly | Tool fields, event order and result return |
| Error handling | Upstream status, error type and request ID can be traced | Gateway error mapping, rate limits and retries |
| Usage reconciliation | Response fields can be reconciled with console billing | Usage field differences, delayed billing and plan rules |
If the current key or model only supports OpenAI-compatible endpoints, test with a compatible client. Do not label the results as native Claude Code compatibility. For environment variables and authentication checks, see the Link Claude Code setup guide, then validate each capability using this checklist. This page does not claim completed tests across all client versions.
Cross-check the Link endpoint and plan group
export ANTHROPIC_BASE_URL="https://www.aifast.link"This sets only the service address, not a key or account permissions. Configure your Link key securely using the setup guide, then restart Claude Code. Do not copy the OpenAI client's /v1 suffix into this variable.
A Claude monthly plan does not cover every Claude group. Compare the plan and current key groups using the Link billing guide. Record daily allowance, wallet fallback and actual charges separately. GPT plans, Claude plans and the separate Gemini group are not interchangeable.
Do not misread common status codes
| phenomenon | Check first | What this does not prove |
|---|---|---|
| 401 | Whether the key is valid, the auth header format and whether the running process loaded the new variable | A 401 alone does not mean the model was removed |
| 403 | Account model access, key restrictions and source policies | A 403 alone does not mean the account was suspended |
| 404 | The final path, version prefix and whether Link docs publish the endpoint | Do not fix a path error by repeatedly changing keys |
| 400 | Client fields, protocol type and server support | Do not treat a Chat Completions request as a Messages request |
| 429 | Current account quota, concurrency limits and Retry-After | Do not retry immediately without limit |
| 502 / 5xx | Request ID, upstream status, timeouts and error body | One failure does not prove long-term unavailability |
Keep the first failure status and request ID, then run one low-cost retest. If only streaming or tool calls fail, locate that protocol capability instead of hiding endpoint incompatibility by changing the account plan.
Link billing and test boundaries
.link Link currently offers GPT, Claude and Gemini models with top-up and monthly subscription options. Before testing, record the plan name, model scope, quota, validity and limits shown in the current console or checkout. Reconcile top-up and subscription results separately. Do not infer quotas, overage handling or Claude Code entitlements that the subscription terms do not state.
Use the following record and fill it again after changing the account plan or model:
Test date:
Site: https://www.aifast.link
Account plan:
Model ID from live catalog:
Protocol documented by Link:
Final request path:
Minimal request / streaming / tool call:
Status and request ID:
Billing record checked:
Unverified capabilities:Continue with AIFast Link OpenAI-Compatible Setup Guide、Link Top-Up and Subscription Cost GuideandModel API comparison test template.