Confidential AI API: Secure LLM Inference at Scale

Beta Router provides a key-aware model catalog. Model availability, context windows, log probabilities, and streaming requirements can differ by API key.

List models available to your key

Call the authenticated Models endpoint with the same API key that you will use for chat requests:

curl --silent --show-error https://router.prem.io/v1/models \
  --header "Authorization: Bearer $PREM_ROUTER_API_KEY"

With the OpenAI Python SDK:

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["PREM_ROUTER_API_KEY"],
    base_url="https://router.prem.io/v1",
)

for model in client.models.list().data:
    print(model.model_dump(exclude_none=True))

The response includes:

id: The public model ID to send in chat requests.
context_length: The context window available to the key.
input_modalities: Supported input types.
capabilities: Feature flags such as tools, vision, and log probabilities.
streaming_required: Present and set to true when the key must use streaming for that model.

Base catalog

The base catalog currently includes six models:

Model ID	Family	Context window	Base64 images	Logprobs	Streaming
`kimi-k3`	Kimi	Up to 1.05M tokens	Yes	No	Optional
`qwen-3.7-max`	Qwen	1M tokens	No	Yes	Key-dependent
`qwen-3.7-plus`	Qwen	1M tokens	Yes	Yes	Key-dependent
`qwen-3.6-plus`	Qwen	1M tokens	Yes	Yes	Key-dependent
`qwen-3.5-9b`	Qwen	262K tokens	Yes	Key-dependent	Optional
`deepseek-v4-pro`	DeepSeek	Up to 1.05M tokens	No	Key-dependent	Optional

Use the exact lowercase model ID returned for your key. Do not add a provider prefix or a slash.

Additional key-dependent models

Some API keys also return these models:

Model ID	Family	Context window	Base64 images	Logprobs	Streaming
`deepseek-v4-flash`	DeepSeek	1.05M tokens	No	Yes	Optional
`qwen-3.8-max`	Qwen	1M tokens	Yes	Yes	Optional

Use an additional model only when authenticated GET /v1/models returns its ID for the same key. qwen-3.8-max requires reasoning to remain enabled. If you change API keys, retrieve the model list again. Remove models that the new key does not return.

Streaming requirements

When a model includes "streaming_required": true, set stream=True in Python or stream: true in JavaScript. If the field is absent, streaming is optional. See API capabilities for tool calls, structured output, reasoning controls, vision, log probabilities, and streamed usage.

Get Started

Router

Core Concepts

Handling Errors

Guides

API Reference

Billing

Support

Models

List models available to your key

Base catalog

Additional key-dependent models

Streaming requirements

​List models available to your key

​Base catalog

​Additional key-dependent models

​Streaming requirements

List models available to your key

Base catalog

Additional key-dependent models

Streaming requirements