Comment by Y_Y

In principle any model can do these. Tool use is just detecting something like "I should run a db query for pattern X" and structured output is even easier, just reject output tokens that don't match the grammar. The only question is how well they're trained, and how well your inference environment takes advantage.