Angelina YangFounder and Host, TwoSetAI · AI builder for visibility and discoverability

Expensive AI Models Overthink

15 Oct 2026 · 1 min

Watch on YouTube

0:42

In this brief dialogue, Speaker A shares findings from a comparative study of state‑of‑the‑art (SOTA) AI models versus smaller open‑source alternatives. The research indicates that the more expensive, high‑capacity models often "overthink" tasks, consuming up to four times more tokens without delivering superior results. Speaker B notes the common belief that higher cost equates to higher quality. Speaker A counters that, in many cases, cheaper models are preferred because they perform adequately for the given task, highlighting that the most expensive model is not always necessary.

I did some research end of last year in comparing some of those SOTA models, and the more expensive ones did overthink literally their task and spent compared to a smaller open-source model, in some cases 4 times as much tokens with not as good results.
People generally believe that things that are more expensive has better quality.
I actually found that a lot of time the cheaper model got preferred because it was just doing fine for that task. So it's not always the most expensive one needs to be overhauled.