Which AI model searches the web better on a KALDI Coffee Farm competitor question?
We asked seven AI models the same question and kept nine answers, two of them from GPT-6.1 Sol at High. Web search is recorded as on for eight of the nine. The record for the ninth does not say. This page counts the websites each answer cites. It does not open the links, so it cannot say whether the pages back what the answers wrote. It compares the models, not their advice.
As asked, word for word.
I want you to research KALDI Coffee Farm (https://www.kaldi.co.jp/) and find the five most similar companies in Japan. Put everything in one comparison table with KALDI as the first row, explain in one line why each company made the list, and link a source for every fact. Write in English.
The question names one company and asks for five more, with one line of reasoning per company and a source link for every fact. Store counts, revenue, and founding dates sit on company pages and in business news, much of it in Japanese. It tests how many sources a model finds and whether it links each fact it states.
Which model researched further?
Show as a table
| Answer | Words | Websites cited |
|---|
More websites means the model looked further. It does not mean the answer is right. That needs the link check below, which has not run for this question.
Do the cited pages back what the models wrote? Not checked yet.
In a checked comparison, we open every cited link and compare the page with the passage that cites it. For this question that has not happened. The numbers below say how much there is to check.
Cited links · one square per link, none opened yet
Until then, treat every figure in these answers as a lead, not a fact. Open the page before a number goes into your plan. To see what a finished check looks like, read the comparison How effective are interpreters and translation tools at trade-show booths in Japan?.
What this means when you pick a model for research.
Every answer names Seijo Ishii.
Seven also name Jupiter. Opus 5.5, Latest at Extra High, and Grok 4.7 say Jupiter filed for civil rehabilitation in January 2026. GPT-6 Luna, both GPT-6.1 Sol answers, and Pro do not mention it.
Two answers drew on different pages.
Opus 5.5 and Sonnet 5.5 never cite kaldi.co.jp. Four sites appear in those two answers and in no other: job.mynavi.jp, scouter.monex.co.jp, bookplus.nikkei.com, and www2.infomart.co.jp. The six other answers that carry links all cite kaldi.co.jp.
Latest at Pro and Grok 4.7 carry few or no links.
The question asked for a link on every fact. Latest at Pro says that every factual claim is supported by cited sources, but its saved copy has no link. Grok 4.7 links two pages, kaldi.co.jp and Wikipedia, and names its other sources in a table column as bare site names, such as meidi-ya.co.jp and yamaya.jp.
Every answer, with its numbers.
| Model | Setting | Asked | Words | Websites | Links |
|---|---|---|---|---|---|
| Opus 5.5 | High | 3 Oct 2026 | 676 | 13 | 16 |
| Sonnet 5.5 | High | 3 Oct 2026 | 599 | 15 | 16 |
| GPT-6 Luna | High | 3 Oct 2026 | 355 | 6 | 9 |
| GPT-6.1 Sol | High | 5 Oct 2026 | 436 | 10 | 12 |
| GPT-6.1 Sol | High | 3 Oct 2026 | 410 | 7 | 13 |
| Latest | Extra High | 4 Oct 2026 | 935 | 8 | 15 |
| Latest | Pro | 4 Oct 2026 | 780 | – | |
| Pro | Pro | 4 Oct 2026 | 547 | 9 | 19 |
| Grok 4.7 | Fast | 5 Oct 2026 | 721 | 2 | 2 |