I wonder how much this affects the model's performance. I imagine Anthropic trains its models to use a generic set of tools, but they can also lean on their spe

I wonder how much this affects the model's performance. I imagine Anthropic trains its models to use a generic set of tools, but they can also lean on their specific tool definitions to save the agent from having to guess which tool for what.