BUG: Default ToolCallingAgent InferenceClient examples failing

[suryabdev]

Problem
Default ToolCallingAgent InferenceClient examples failing
TLDR: The nscale provider does not support tool calling

Steps to reproduce
The following code fails with

from smolagents import ToolCallingAgent, InferenceClientModel, WebSearchTool

model = InferenceClientModel()

agent = ToolCallingAgent(tools=[WebSearchTool()], model=model)
agent.run("Could you get me the title of the page at url 'https://huggingface.co/blog'?")

with

Bad request:
{'code': '422', 'error_type': 'UNSUPPORTED_OPENAI_PARAMS', 'message': 'The following parameters are not supported for this model: tools, tool_choice', 'param': 'tools, tool_choice'}

Actual behavior and error logs
The same code was working fine yesterday, so then I thought it might be an issue with the provider. On a deeper look

# Fails
model = InferenceClientModel(provider="nscale")
# But the following works
model = InferenceClientModel(provider="together")

I took a look at the Inference documentation. I think HF Inference API automatically picks the cheapest one for usage, nscale would be selected now (Which does not support Tool calling). But together might have been selected till yesterday

Expected behavior
The default parameters for InferenceClient should work

Bottom line, The default parameters will break ToolCallingAgent. Should we change the InferenceClient default to a similar model like Qwen/Qwen3-Coder-30B-A3B-Instruct?

smolagents/src/smolagents/models.py

Line 1401 in f042c0c

model_id: str = "Qwen/Qwen2.5-Coder-32B-Instruct",

All Qwen3-Coder-30B-A3B-Instruct providers support ToolCalling, and the Qwen3-Coder models are better for code generation and tool calling tasks. Price is similar too

Environment:
Please complete the following information:

• OS: Linux / WSL
• Python version: 3.12.3
• Package version: 1.22

Full error traceback

---

Checklist

• I have searched the existing issues and have not found a similar bug report.
• I have provided a minimal, reproducible example.
• I have provided the full traceback of the error.
• I have provided my environment details.
• I am willing to work on this issue and submit a pull request. (optional)

[aymeric-roucher]

Good point @suryabdev ! Indeed I think we need to update models used as defaults and in the doc I did some updates in #1801 but we need a bigger one across the board.

[suryabdev]

I can update the rest in a follow up PR to #1801, I'll start off with Qwen/Qwen3-Coder-30B-A3B-Instruct as the replacement

[suryabdev]

Raised #1813 to change the default

[suryabdev]

#1813 has been merged, Closing this issue