Available Models

Mume Gateway gives you access to 500+ models from the world's leading AI providers through a single API. Use the provider/model format when specifying models.

Model Format

All models follow the provider/model-name convention. For example:

Model Names

anthropic/claude-haiku-4.5
anthropic/claude-opus-5
google/gemini-3.5-flash
mistralai/voxtral-small-24b-2507
moonshotai/kimi-k2-thinking

Providers

Provider	Example Models
OpenAI	openai/gpt-4.1-mini, openai/gpt-4.1-nano, openai/gpt-5.1
Anthropic	anthropic/claude-opus-5, anthropic/claude-sonnet-5, anthropic/claude-haiku-4.5
Google	google/gemini-3.5-flash
Mistral	mistralai/voxtral-small-24b-2507
Moonshot	moonshotai/kimi-k2-thinking
Meta	meta-llama/llama-4-maverick
DeepSeek	deepseek/deepseek-chat
xAI	x-ai/grok-4.5

Usage

Python

# Use any model by specifying provider/model-name
response = client.chat.completions.create(
    model="anthropic/claude-haiku-4.5",
    messages=[{"role": "user", "content": "Hello!"}],
)

# Switch providers instantly — same API, same code
response = client.chat.completions.create(
    model="google/gemini-3.5-flash",
    messages=[{"role": "user", "content": "Hello!"}],
)

GET /models

The catalog is an API, and it is public — no key required. It is a price list and a capability list for models anyone can read about on the marketing pages, so a token bought nothing.

Bash

curl https://mume.ai/api/v1/models

JSON

{
  "object": "list",
  "data": [
    {
      "id": "anthropic/claude-haiku-4.5",
      "object": "model",
      "created": 1760547638,
      "owned_by": "anthropic",
      "name": "Claude Haiku 4.5",
      "description": "Claude Haiku 4.5 is Anthropic's fastest and most efficient model…",
      "input_modalities": ["image", "text"],
      "output_modalities": ["text"],
      "context_length": 200000,
      "surface": "chat",
      "pricing": {
        "prompt_per_million": 1,
        "completion_per_million": 5,
        "currency": "USD"
      },
      "supported_parameters": ["tools", "tool_choice", "reasoning", "…"],
      "intelligence_index": 30,
      "intelligence_basis": "artificial-analysis"
    }
  ]
}

Narrowing the list

Two query parameters, and they answer different questions. output filters on what a model produces; surface filters on which endpoint invokes it.

Bash

# every model that can return video
curl "https://mume.ai/api/v1/models?output=video"

# everything you can POST to /audio/speech
curl "https://mume.ai/api/v1/models?surface=speech"

The distinction matters for audio: a speech model and a transcription model each report their own modality, so outputcannot express "things the speech endpoint accepts". surface can. The values in the catalog today are chat, media, speech, transcription and embeddings — note that image and video models are both media, because both are invoked the same way.

Cache it

The full response is ~350 KB uncompressed for 500+ models, and the catalog only moves when our sync runs. It is sent cache-control: public, max-age=300 with an ETag, and a conditional request is answered with a 304 and zero bytes:

Bash

curl -H 'If-None-Match: W/"57018-410fNFbltOeCc74bv0MHJBT8d9g"' \
  https://mume.ai/api/v1/models
# HTTP/2 304

Fetch it once at start-up and revalidate rather than calling it per request. It is frequently the first call an integration makes and the easiest one to make far too often.

GET /models/{id}

One model, by the same id model accepts. The id contains a slash and does not need encoding.

Bash

curl https://mume.ai/api/v1/models/anthropic/claude-haiku-4.5

The body is a single entry in the shape above — not wrapped in a list. An unknown id is a 404 with code: "model_not_found", which is the cheapest way to validate a model id a user typed before you spend a completion on it.

GET /models/{id}/providers

Who can serve a given model, cheapest first — context window, tool support, price, and measured latency and throughput per provider.

Bash

curl https://mume.ai/api/v1/models/anthropic/claude-haiku-4.5/providers

JSON

{
  "object": "list",
  "data": [
    {
      "id": "google-vertex/global",
      "name": "Google",
      "context_length": 200000,
      "max_completion_tokens": 64000,
      "supports_tools": true,
      "pricing": { "prompt_per_million": 1, "completion_per_million": 5, "currency": "USD" },
      "latency_p50_ms": 477,
      "throughput_p50_tps": 76
    }
  ]
}

An empty list is a 200, not an error.Upstream being unreachable means we do not know the options, and the honest response to that is to offer none and leave routing to us — not to fail your page. This one is cached for 60 seconds rather than 300, because it carries live latency and a stale "fastest provider" is worse than a slightly slower answer.

Browsing Models

If you would rather look than parse, the same catalog is on the Models page, filterable by provider, capability and price.

Image Generation

The developer API also supports image generation models via the images endpoint:

Endpoint

POST https://mume.ai/api/v1/images/generations

← Web Search Next: Error Handling →