mistralai/mistral-nemo doesn’t return tool calls?

[fading gust]

Hi all I’m using openrouter/auto which seems to be sending me to the mistral model.

I’m using OpenAI SDK in python and registering tools. It works with openai/gpt-oss-20b for example, but when I land on mistral I get tool calls within the text output like this:

Plan: Since the menu is already detected, the goal is complete.

Tools:
- `ReportGoalAchieved`: {\"confidence\": \"high\", \"reasoning\": \"Menu detection was successful.\"}",
  "goal_achieved": false,
  "tool_calls": []
}

Shouldn’t auto be routing me to a model that supports tools?

[runic rampart]

I'm an AI assistant - my answers may be incorrect and I cannot see screenshots. Please verify information with the OpenRouter docs.

The OpenRouter Auto Router (openrouter/auto) dynamically selects from a pool of models, including Mistral models like mistralai/mistral-nemo. However, not all models in the pool support OpenAI-style tool/function calling. The Mistral models, including mistralai/mistral-nemo, generally do not support proper tool calls and may instead output tool calls as plain text, as you're seeing.

"Requests will be routed to the following models: ... mistralai/mistral-nemo ..."
But there is no explicit indication that mistralai/mistral-nemo supports tools, and it is common for non-OpenAI models to lack full tool/function calling compatibility OpenRouter.

If you require tool calling, you should explicitly specify a model that supports it (like openai/gpt-oss-20b or openai/gpt-4o). The Auto Router may select models that do not, unless you restrict the candidate models.

For more on tool calling and supported models:
"Use tools (or functions) in your prompts with OpenRouter. Learn how to use tools with OpenAI, Anthropic, and other models that support tool calling."
See the Tool & Function Calling documentation.

To ensure tool calling works, specify only models you know support it using the models parameter (not just openrouter/auto). If you need help configuring this, let me know:

• What exact error or output do you see when tools fail?
• Which model is listed in the response's model field when it fails?
• Can you share your relevant code snippet for how you're registe...