Supported providers
| Provider | Models | Description |
|---|---|---|
| Anthropic | Claude Opus 4.6, Sonnet 4.6, Haiku 4.5 | Claude models with extended thinking support |
| OpenAI | GPT-5.4, GPT-5.3 Codex, GPT-5.3 Instant | GPT and Codex models |
| Gemini 3.1 Pro, Gemini 3 Flash, Gemini 3.1 Flash Lite | Gemini models with large context windows | |
| Moonshot AI | Kimi K2.5 | Multimodal dialogue and agent tasks |
| Alibaba | Qwen 3-32B, Qwen 3-Coder 30B-A3B | Cost-effective coding models |
Endpoint modes
Each provider can be configured with one of three endpoint modes:stagewise (default)
Requests are routed through the stagewise proxy. This is the simplest setup — no API key required. Usage is billed through your stagewise account.Official API
Connect directly to the provider’s official API endpoint. You need to supply your own API key. This is the standard BYOK (Bring Your Own Key) setup. See the BYOK setup guide for step-by-step instructions.Custom endpoint
Route requests to a custom endpoint that implements one of the supported API specifications. Useful for:- Self-hosted models (e.g. vLLM, Ollama)
- Enterprise proxy servers
- Alternative API providers (Azure, AWS Bedrock, Google Vertex)
Model capabilities
Each model declares its supported input and output modalities:- Text — All models support text input and output
- Image — Most models accept image attachments (JPEG, PNG, GIF, WebP)
- File — PDF file uploads (supported by most models)
- Video — Video input (Google and Moonshot models only)
- Tool calling — All built-in models support function/tool calling
Input size limits vary by provider. For example, Anthropic limits images to 5 MB while Google allows up to 100 MB.