How to connect
Get an API key from the provider's platform and pay by usage. You can also add a custom provider by hand.
Built-in providers
| Provider | Access | Built-in models |
|---|---|---|
| DeepSeek | API key | DeepSeek V4 Pro, DeepSeek V4 Flash |
| Kimi (Moonshot) | API key | Kimi K3, Kimi K2.7 Code |
| Kimi Code (Coding Plan) | API key | Kimi K2.7, Kimi For Coding |
| Zhipu BigModel (GLM) | API key | GLM 5.2, GLM 5.1, GLM 5 Turbo, GLM 4.7 FlashX |
| Zhipu Code (Coding Plan) | API key | GLM 5.2, GLM 5.1, GLM 5 Turbo, GLM 4.7, GLM 4.5-Air |
| MiniMax (mainland China) | API key | MiniMax M3, M2.7 (incl. Highspeed), M2.5 (incl. Highspeed) |
| Xiaomi MiMo | API key | MiMo V2.5 Pro, MiMo V2.5 |
Fill in a base URL and key for a custom provider, and Finch will detect the API type and fetch the model list automatically.
Quick vs. Think
Every conversation can run in Quick or Think mode: Quick answers right away without much deliberation, for the fastest response; Think reasons before answering, is more careful, and lets you tune how deep it thinks (light / moderate / careful / deep). Not every model supports thinking — it depends on whether the underlying model has that capability.
Different models for different tasks
Everyday chats and automation tasks can each use their own model and thinking level — for example, your best flagship model for daily use, and a fast, lightweight model for scheduled automation.