Prompt
Which LLM API providers support function calling and strict JSON mode for automation workflows?
Latest observation
Jul 19, 2026
OpenAI APIWeb search: off
Several major LLM API providers support function calling and some form of strict JSON / structured output for automation workflows. Here’s a practical list:
Providers with strong support
1) OpenAI
- Function calling / tool calling: Yes
- Strict JSON / structured outputs: Yes
- Best known for reliable structured outputs and agent/tool workflows.
2) Anthropic
- Function calling / tool use: Yes
- Strict JSON / structured outputs: Supported via tool use and prompting patterns; exact “strict JSON mode” is less native than OpenAI’s structured outputs, but usable for automation.
3) Google Gemini API
- Function calling: Yes
- JSON output / schema-based output: Yes, supports structured responses and tool use.
- Good option for workflows needing integration with Google ecosystem.
4) Mistral API
- Function calling / tool calling: Yes
- JSON mode / structured outputs: Supported in newer API features and model variants.
- Useful if you want a lightweight alternative.
5) Cohere
- Tool use / function calling: Yes
- Structured output: Available in their API offerings, depending on model and endpoint.
6) xAI Grok API
- Function calling/tool use: Available in supported API versions.
- Structured output: Limited compared with OpenAI/Gemini, but improving.
Also commonly used through compatible APIs
Some providers expose OpenAI-compatible endpoints, which often make it easier to use function calling patterns:
- Together AI
- Fireworks AI
- Groq
- DeepSeek
- OpenRouter (as a router/aggregator rather than a model provider)
Support for “strict JSON” varies by model and endpoint, so you need to verify per model.
What to look for
For automation workflows, the key features are:
- Tool/function calling
- Schema-constrained output
- JSON mode / response format
- Deterministic retries / validation
- Streaming support if you need low latency
Best practical choices
If you want the most dependable setup today:
- OpenAI — strongest structured-output tooling
- Google Gemini — good structured output + function calling
- Anthropic — excellent tool use, though structured JSON is usually less “strict” natively
- Mistral — solid if you want open-ish deployment options
If you want, I can also give you a comparison table with:
- provider
- function calling support
- strict JSON support
- schema enforcement
- open-source/open-weight options
- pricing/latency notes