You're both completely missing the point. It's important that an LLM be able to perform exact arithmetic reliably without a tool call. Of course the underlying hardware does so extremely rapidly, that's not the point.
The computer ALREADY does do math reliably. You are missing the point.
Could you explain why that is?