Skip to main content
This guide walks through using Anthropic Claude models with the AI Gateway. Anthropic is a built-in provider you can pay for with credits, so you can use Claude models without an Anthropic account of your own, or bring your own key.

Pay with credits (no Anthropic account)

You can use Claude models without your own Anthropic account. ngrok.ai supplies the Anthropic credentials and bills the model cost to your credits.
1

Create an access key and add credits

Follow the quickstart to create an access key and purchase credits in app.ngrok.ai.
2

Send requests

Point your SDK at the gateway with your access key:
The gateway supports both the Anthropic Messages API (/v1/messages) and OpenAI-compatible Chat Completions.

Bring your own Anthropic key

Want requests billed to your own Anthropic account instead of through credits? Store your Anthropic key once—the gateway attaches it for you on every request.
1

Create an access key

Follow the quickstart to create an access key in app.ngrok.ai if you don’t have one yet.
2

Add your Anthropic key

Add a provider key for Anthropic. ngrok stores it encrypted, so it never has to live in your application or get shared around—and you can reuse it across as many access keys as you need.
3

Route to it with a configuration

Create an access key configuration with an Anthropic routing rule set to Bring your own API key, then assign it to your access key.
4

Keep sending your access key

Your code doesn’t change. Send the same access key as before—the gateway swaps in your Anthropic key before the request leaves for Anthropic.

Explicit provider routing

Available models

Browse the model catalog for the full list of Claude models you can call through the gateway.

Parameters the gateway removes

The gateway removes these fields before it sends a request to Anthropic, and lists them in the Ngrok-AIG-Removed-Params response header:
  • speed on every Claude model. Fast mode isn’t available through the gateway, so a request with "speed": "fast" runs and bills at standard speed.
  • temperature, top_p, and top_k on the models that reject them: the Claude Fable models, Claude Opus 4.7 and later, Claude Sonnet 5 and later, and Claude Haiku 5.5.
  • reasoning_effort on Chat Completions requests to every Claude model. To set the effort level, send an Anthropic Messages request to /v1/messages with output_config, which the gateway forwards.

Fable 5.2

Claude Fable 5.2 (claude-fable-5-2) has a 1M-token context window, up to 128K output tokens, and always uses adaptive thinking. It takes text, image, and file input. Effort is the main control for this model, so call it with the Messages API. The gateway removes reasoning_effort from Chat Completions requests. Anthropic returns a 400 error for some Claude Fable 5.2 requests that Claude Fable 5.1 accepts, and the gateway passes the error through. Anthropic rejects a request that:
  • Forces tool use with tool_choice.
  • Sets thinking to enabled or disabled.
  • Uses the computer_20251124 tool instead of computer_toolset_20260801.
  • Edits the system prompt, the tools, or an earlier message, and then sends back a Claude Fable 5.2 thinking block.

Files

Anthropic also supports the Files API for uploading documents and images once and referencing them by ID.

Next steps