Builder reference
Model API compatibility at a glance
These rows are intentionally approximate. Always confirm tool-calling, JSON mode, and limits in the provider’s docs for the exact model id you deploy. Use this table to plan migrations and Checks coverage.
| Provider | Tool calling | JSON / structured | Context (approx.) | Notes |
|---|---|---|---|---|
| OpenAI | native | native | Varies by model (e.g. 128k–1M class) | Responses + Chat Completions; JSON/schema modes differ by endpoint. |
| Anthropic | native | partial | Long context on Sonnet/Opus tiers | Tool use is first-class; structured output patterns vary by SDK. |
| Google (Gemini) | native | native | Large context on 1.5/2 family | Vertex vs AI Studio feature parity can differ. |
| Meta (Llama) | varies | varies | Depends on host and fine-tune | Open weights: behavior depends on inference stack and chat template. |
| Mistral | native | partial | Model-dependent | Check function-calling docs for your specific model id. |
| Azure OpenAI | native | native | Matches OpenAI model deployment | Enterprise controls, regions, and content filters affect routing. |
| Cohere | native | partial | Model-dependent | Command R+ and tool-use APIs differ from embed/classify endpoints. |
| Perplexity | varies | varies | Product-dependent (API vs consumer) | Confirm structured output and tools on the exact API product you use. |