Developer prices pairwise LLM ticket dedup, costs scale quadratically

On November 13, 2026 at 20:00, the AI section of DEV Community published an analysis of what it costs to use an LLM for pairwise comparison of support tickets, the standard approach for flagging tickets that describe the same underlying problem.

The article explains that LLMs handle this kind of matching well. Present two tickets together, ask whether they refer to the same issue and request a yes-or-no answer, and the model usually gets it right. A common implementation is therefore a loop in which every ticket is compared against every other ticket, once each.

The article's cost estimates are roughly $111 for 1000 tickets and $11130 for 10000 tickets. It attributes this growth to the quadratic scaling of the approach: N tickets generate N×(N−1)/2 model calls, and since Model APIs bill per token, input tokens dominate the bill, with a single input text averaging about three-quarters of an English word in length. The author notes that the approach costs about a dollar on 100 test tickets, so it is easy to ship, but once it runs against a real backlog of accumulated tickets, the bill climbs faster than the data volume itself.

No follow-up reporting has appeared in the available sources, so the discussion currently stands at this single published analysis with no further developments to report.