You are probably good with any leading model, but yeah I'd reach for Llama myself if I were doing some cover letters or similar work rn. From my experience, using models like Llama to cough out boilerplate and menial writing for job applications and the like. It cuts down on the uselessly slow parts of a lot of that part but-- My advice is not to use the entire enerated cover letter but to copy the text to another doc and try and re-write 20-30% and cut down as much filler words and sentences as possible.

Then, Read it out loud to hear if it's still too robotic or inhuman and adjust as needed to make it sound like a real person (which is why reading it out loud is important-- grammar slip-ups, typos, awkward or confusing wordings/sentences are easily glossed over when proofreading silently and quickly but caught instantly when you hear them spoken/ speak/ etc..)

You end up avoiding the risk of any overly AI-rididity by adding just enough of your humanity by rewording places that have sentences that you personally would never say (in that way or at all). 

Works well for using AI for rote/redundant things that otherwise feel way too time consuming for what it's worth (example: Cover letters that aren't just sorta plug-and-play madlibs but generated by AI to hit all the Job description's salient points specifically/adeptly while saving you enough time to proofread and re-phrase/adjust as needed to ensure there's some trace of you there and not just the notorious GPT pet phrases and wordings that are jarring to any human who reads the letter.)