The 47 Point Cliff

追加: 2026-05-16

[00:00:00]Eight AI models scored between 91% and 93% on our Agent e-commerce benchmark.

[00:00:07]Then, there's a 47-point cliff.

[00:00:10]Kimi K 2.6 scored 44%.

[00:00:14]Gemini 3.1 Pro scored 36%.

[00:00:18]Gemini picked the wrong product 75% of the time.

[00:00:23]Google just added buy now, pay later to Gemini shopping.

[00:00:27]Klarna and Affirm built right in.

[00:00:31]The model that picks the wrong product three out of four times can now also finance the wrong product for you.

[00:00:38]Full data in the newsletter at tabverified.substack.com.

[00:00:42]Not vibes, verified. tabverified.ai