Model Catalog - ngrok documentation

ngrok.ai maintains a catalog of known providers and models. This catalog enables automatic model resolution, routing, model validation, and rich metadata for selection strategies.

How the catalog works

When you send a request with a model name like gpt-4o, the gateway automatically:

Resolves the model using the catalog
Finds every provider that serves the model, with the model author’s own provider (for example, OpenAI) first, and applies the right provider credentials
Routes the request to the provider’s endpoint

To explicitly select a provider, use the provider:model format in the model string. For example:

{
  "model": "openai:gpt-4o",
  "messages": [{"role": "user", "content": "Hello"}]
}

Built-in providers

The catalog includes built-in providers—public AI APIs with known models and routing metadata. You can manage them on the Providers page.

Authentication

Send traffic using an access key. This key authenticates your requests to the gateway, not to the upstream provider.

Billing and provider credentials

The AI Gateway splits billing into two parts: using the gateway and paying the model provider. Many built-in models don’t need a provider key. You can spot them in the tables below: if Needs BYOK? says “No,” you can use the model right away as long as you have credits. The gateway takes care of authentication and charges your credits. Availability is per model, so a provider can have some models you can pay for with credits and some you can’t. Other providers do need a key. For those, add your provider key in app.ngrok.ai. The provider bills you directly for model usage, and your credits cover the gateway fee. This is often called BYOK (bring your own key).

OpenAI

Field	Value
Provider ID	`openai`
Base URL	`https://api.openai.com/v1/`
Website	openai.com
API surfaces	Responses, Chat Completions, Decisions, Images
Provider key required	No

How to use OpenAI →

OpenAI models

Model ID	Context	Max Output	Modalities

gpt-image-2, gpt-image-2.5-sunburst, and gpt-image-2.5-flare generate and edit images. Send them to the Images API at /v1/images/generations or /v1/images/edits, not to Chat Completions or Responses.

Anthropic

Field	Value
Provider ID	`anthropic`
Base URL	`https://api.anthropic.com/v1/`
Website	anthropic.com
API surfaces	Chat Completions, Messages
Provider key required	No

How to use Anthropic →

Anthropic models

Model ID	Context	Max Output	Modalities

Anthropic retires claude-sonnet-4-5 on November 30, 2026. Move requests to claude-sonnet-5-5 before then. Anthropic returns errors for some Claude Fable 5.2 requests that Claude Fable 5.1 accepts. See Fable 5.2.

Google

Field	Value
Provider ID	`google`
Base URL	`https://generativelanguage.googleapis.com/v1beta/openai/`
Website	aistudio.google.com
API surfaces	Chat Completions, Gemini GenerateContent, Interactions
Provider key required	For some models. See Needs BYOK?

How to use Google →

Google models

Model ID	Context	Max Output	Modalities

gemini-3-pro-image, gemini-3.1-flash-image, gemini-3.1-flash-lite-image, and gemini-nano-banana-2.1 return images as well as text. They serve Gemini GenerateContent and Interactions requests, not Chat Completions. Their Context column shows the input limit. Google rejects image input to the Gemma 4 models, so the gateway accepts only text for them on Google.

Google serves the Gemini 2.5 models only to Google projects created before mid-2026. Newer projects get a 404 stating the model is no longer available to new users. The Google project ngrok.ai uses postdates that cutoff, so these models are reachable only with your own key on an older project.

DeepSeek

Field	Value
Provider ID	`deepseek`
Base URL	`https://api.deepseek.com/v1`
Website	deepseek.com
API surfaces	Chat Completions, Responses, Messages
Provider key required	No

How to use DeepSeek → When you pay with credits, DeepSeek models bill at DeepSeek’s off-peak price at every hour. See Peak and off-peak pricing.

DeepSeek models

Model ID	Context	Max Output	Modalities

xAI

Field	Value
Provider ID	`xai`
Base URL	`https://api.x.ai/v1`
Website	x.ai
API surfaces	Chat Completions, Responses, Images
Provider key required	No

How to use xAI →

xAI models

Model ID	Context	Max Output	Modalities

grok-imagine-image and grok-imagine-image-2.0 generate and edit images. Send them to the Images API at /v1/images/generations or /v1/images/edits.

Groq

Field	Value
Provider ID	`groq`
Base URL	`https://api.groq.com/openai/v1`
Website	groq.com
API surfaces	Chat Completions, Responses
Provider key required	Yes

How to use Groq → Groq provides AI inference powered by their custom LPU (Language Processing Unit) hardware.

Groq models

Model ID	Context	Max Output	Modalities

Mistral AI

Field	Value
Provider ID	`mistral`
Base URL	`https://api.mistral.ai/v1`
Website	mistral.ai
API surfaces	Chat Completions
Provider key required	Yes

How to use Mistral AI →

Mistral AI models

Model ID	Context	Max Output	Modalities

Moonshot AI

Field	Value
Provider ID	`moonshotai`
Base URL	`https://api.moonshot.ai/v1`
Website	moonshot.ai
API surfaces	Chat Completions
Provider key required	No

How to use Moonshot AI →

Moonshot AI models

Model ID	Context	Max Output	Modalities

Z.ai

Field	Value
Provider ID	`z-ai`
Base URL	`https://api.z.ai/api/paas/v4`
Website	z.ai/model-api
API surfaces	Chat Completions, Images
Provider key required	No

How to use Z.ai →

Z.ai models

Model ID	Context	Max Output	Modalities

glm-image generates images. Send it to the Images API at /v1/images/generations. It doesn’t edit images.

QwenCloud

Field	Value
Provider ID	`qwencloud`
Base URL	`https://dashscope-intl.aliyuncs.com/compatible-mode/v1`
Website	qwencloud.com
API surfaces	Chat Completions, Messages, Responses
Provider key required	No

How to use QwenCloud →

QwenCloud models

Model ID	Context	Max Output	Modalities

MiniMax

Field	Value
Provider ID	`minimax`
Base URL	`https://api.minimax.io/v1`
Website	minimax.io
API surfaces	Chat Completions, Responses, Messages
Provider key required	No

How to use MiniMax →

MiniMax models

Model ID	Context	Max Output	Modalities

MiniMax retires minimax-m2.7, minimax-m2.7-highspeed, minimax-m2.5, and minimax-m2.5-highspeed on November 18, 2026. Move requests to minimax-m3 or minimax-m3.1-flash before then.

Tencent TokenHub

Field	Value
Provider ID	`tokenhub`
Base URL	`https://tokenhub-intl.tencentcloudmaas.com/v1`
Website	tencentcloud.com/document/product/1300
API surfaces	Chat Completions, Responses, Messages
Provider key required	No

How to use Tencent TokenHub →

Tencent TokenHub models

Model ID	Context	Max Output	Modalities

TokenHub retires two models:

minimax/minimax-m2.7 on November 17, 2026, at 16:00 UTC. Move requests to minimax/minimax-m3.
tencent/hy4-preview on December 6, 2026, at 16:00 UTC. Move requests to tencent/hy4.

TokenHub has also deprecated deepseek/deepseek-v4-flash. It still answers, but ngrok/auto no longer picks it. Use deepseek/deepseek-flash in new code.

Typesafe AI

Field	Value
Provider ID	`typesafeai`
Base URL	`https://api.typesafe.ai/v1`
Website	typesafe.ai
API surfaces	System One
Provider key required	No

How to use Typesafe AI →

Typesafe AI models

Model ID	Context	Max Output	Modalities

Liquid AI

Field	Value
Provider ID	`liquid`
Base URL	`https://api.liquid.ai/openai/v1`
Website	liquid.ai
API surfaces	Chat Completions, System One
Provider key required	No

How to use Liquid AI →

Liquid AI models

Model ID	Context	Max Output	Modalities

The LFM models serve Chat Completions. The d1 models are decision models that serve System One requests at /v1/systemone, and their Context column shows the input limit.

LLM API

Field	Value
Provider ID	`llmapi`
Base URL	`https://api.llmapi.ai/v1`
Website	llmapi.ai
API surfaces	Chat Completions, Responses, Messages, Images, System One
Provider key required	For some models. See Needs BYOK?

How to use LLM API → LLM API serves models from many labs under one API key. Many of them are also served by their own labs, so add the llmapi: prefix, as in llmapi:google/gemma-4-31b-it, to keep a request on LLM API.

LLM API models

Model ID	Context	Max Output	Modalities

LLM API can send a request to a different upstream than the one the model names, and it bills that upstream’s price. A model works with credits only when no upstream charges more than the catalog price. Models with long-prompt price tiers, separate cache-write prices, or image input priced apart from text need your own LLM API key, and that includes every Claude model. See LLM API for the request formats each model accepts.

OpenRouter

Field	Value
Provider ID	`openrouter`
Base URL	`https://openrouter.ai/api/v1`
Website	openrouter.ai
API surfaces	Responses, Chat Completions, Messages
Provider key required	Yes

How to use OpenRouter → OpenRouter is a unified API that provides access to multiple AI models from various providers through a single endpoint. The catalog lists no models for OpenRouter. Send any model ID the provider serves with the openrouter: prefix, and the gateway passes it through.

Hyperbolic

Field	Value
Provider ID	`hyperbolic`
Base URL	`https://api.hyperbolic.xyz/v1`
Website	hyperbolic.ai
API surfaces	Chat Completions
Provider key required	Yes

How to use Hyperbolic → Hyperbolic provides high-performance inference for open-source models. The catalog lists no models for Hyperbolic. Send any model ID the provider serves with the hyperbolic: prefix, and the gateway passes it through.

InceptionLabs

Field	Value
Provider ID	`inceptionlabs`
Base URL	`https://api.inceptionlabs.ai/v1`
Website	inceptionlabs.ai
API surfaces	Chat Completions
Provider key required	Yes

How to use InceptionLabs → InceptionLabs develops diffusion-based language models for fast, efficient text generation. The catalog lists no models for InceptionLabs. Send any model ID the provider serves with the inceptionlabs: prefix, and the gateway passes it through.

Inference.net

Field	Value
Provider ID	`inference.net`
Base URL	`https://api.inference.net/v1`
Website	inference.net
API surfaces	Chat Completions
Provider key required	Yes

How to use Inference.net → Inference.net provides a distributed inference network for running AI models at scale. The catalog lists no models for Inference.net. Send any model ID the provider serves with the inference.net: prefix, and the gateway passes it through.

Using models from the catalog

Simple model reference

Reference models directly by their ID. A bare ID such as gpt-4o resolves to every provider that serves the model. The full catalog ID, such as openai/gpt-4o, works too:

{
  "model": "gpt-4o",
  "messages": [{"role": "user", "content": "Hello"}]
}

Explicit provider

Use the provider:model format for explicit routing:

{
  "model": "openai:gpt-4o",
  "messages": [{"role": "user", "content": "Hello"}]
}

Custom providers

Built-in providers cover major public APIs. For self-hosted models (Ollama, vLLM, LM Studio) or other endpoints, create a custom provider in the Custom tab in app.ngrok.ai. Custom providers always require a provider key when the upstream needs authentication. See Custom providers to add one.

Catalog updates

ngrok.ai updates the model catalog periodically to include new models and providers. The tables on this page leave out models that their provider has retired. To use a model that isn’t in the catalog yet, send it with its provider prefix, such as groq:new-model-id. The gateway passes it through to that provider. Uncataloged models need your own provider key, because ngrok.ai can’t price them, so you can’t pay for them with credits.