https://www.perplexity.ai/ is trying to do this by combining GPT-3 and Bing search results. Asking a question like "when did Abraham Lincoln land on the moon?" reveals there's a lot of work to be done here.

As far as I know no one has a good language model that can attribute its responses yet.

Edit: I reread what you wrote and I realized you're looking specifically for peer reviewed sources. I'm not sure such a system exists yet. Galactica by Meta might be the closest, but it was only trained over a collection of academic sources that had appropriate licensing terms.