Models
How CanalAPI exposes hundreds of models behind a single endpoint.
CanalAPI aggregates models from multiple providers behind a single OpenAI-compatible endpoint. You select the model in each request via the model field, and the platform routes to the appropriate upstream.
Listing models
There are three authoritative sources:
- Console → Models (signed-in; filter by price / context / capability)
- Public page /models (no sign-in; shows the publicly visible catalog)
- REST endpoint
GET /v1/models— programmatic listing of models available to your account
Each entry includes:
- The canonical name to put in the
modelfield (one-click copy) - Provider and family
- Category: Chat / Reasoning / Image / Audio
- Context window (max tokens)
- Capabilities (text / vision / tools / reasoning / multimodal)
- Input / output / cache pricing (per million tokens)
Refer to the console for the latest names, capabilities, and prices. The list evolves as providers ship new models and the platform onboards them.
Selecting a model
Set the model field in the request body:
{
"model": "gpt-4o-mini",
"messages": [{"role": "user", "content": "Hello"}]
}Model families
CanalAPI currently integrates the following families (actual SKUs depend on the console):
| Category | Providers / families |
|---|---|
| Overseas flagships | OpenAI (GPT-4o / o-series reasoning), Anthropic (Claude), Google (Gemini), xAI (Grok) |
| Chinese majors | DeepSeek, Moonshot / Kimi, Zhipu GLM, Alibaba Qwen, Baidu ERNIE, MiniMax |
| Multimodal | Vision variants (GPT-4o, Claude Sonnet, Gemini Pro, Qwen-VL, …), image generation (DALL·E, …), audio (Whisper / TTS) |
Choosing the right model
| Scenario | Suggested |
|---|---|
| High-volume classification, summarization, routing | Smaller, faster, cheaper models (e.g. *-mini, lightweight open models) |
| Conversational assistant, general-purpose use | Mid-tier flagships (GPT-4o, Claude Sonnet, Gemini Pro, Qwen Max) |
| Complex reasoning, code synthesis | Top-tier or dedicated reasoning models (OpenAI o-series, Claude Opus, DeepSeek-R1-family) |
| Vision input | Models with documented vision capability |
| Long context (100k+ tokens) | Models with sufficient context window (Gemini 1.5/2, Claude family) |
| Chinese-heavy workloads | Chinese families usually offer better quality-per-dollar on Chinese text |
Protocol compatibility
Most models are reachable through any of the three protocols (see Endpoints):
- OpenAI Chat Completions / Responses (recommended)
- Anthropic Messages (native Claude)
- Gemini native (
generateContent/streamGenerateContent)
Using the same model name under any protocol routes to the same upstream model; quotas and pricing are computed per account regardless of protocol.
Compatibility notes
- Provider-specific fields that don't fit the OpenAI shape are passed through as extension fields.
- Quality, latency, and price differ substantially between models for nominally similar tasks; benchmark with your own evaluation set before committing.
- Some models bill tool calls, vision input, or cached prefixes separately — check the model's detail page in the console.