• This thread is just the tip of the iceberg.The people ahead of the curve aren't Googling for answers — they're already in here, having the conversations you haven't found yet. DealerRefresh is free.Get the full picture →

Remember when AI was $$$ CHEAP?

Joe from the product side it's the opposite of the seat cost you're seeing. Unit prices keep falling. A background swap on a listing photo runs us about a penny all in, classify plus edit plus upscale. Video is the one that still costs real money per car so we meter that one.

Where the bill actually lives is the interactive work. The reasoning tokens for the agent that helps us build the product cost more per month than every image we generate for every dealer put together. Alex is right that Sonnet class handles most of the retail data work. We only run the big models on the parts that touch a customer or money and route everything else to the cheap ones. Model routing is most of the cost control at this point.
 

✨ AI Highlights

Dealers are noticing AI token limits tightening and costs rising, drawing comparisons to the early days of cheap PPC clicks. The thread debates whether running AI locally is a viable cost-saving alternative, with one contributor noting that mid-tier open-source models are capable enough for automotive retail tasks, while another cautions that local hardware costs and a six-month capability gap behind frontier models complicate the math. A late reply notes that a newer model version dropped with better price-to-performance ratios, though usage bills are still climbing.

Replies Views 10 976 Started Last Reply