Content density chart of this comparison: words in each answer against websites cited
← Library · Model comparisons Model comparison · web search

Which AI model searches the web better on a Japan e-commerce cost question?

We gave seven AI models the same question about e-commerce channels in Japan for overseas food providers. Haiku 5.5 answered at two settings, so there are eight answers. This page measures how far each answer searched. Its cited links are not checked yet, so it does not say whether the pages back what the models wrote. It compares the models, not their advice.

AI-generated · unverified · asked 8 Oct 2026 · links not checked

The test question

As asked, word for word.

What are the top 10 e-commerce channel for overseas food providers in Japan. We have limited budget, but we want to do more than testing.

My goal here is not to compare superficial price and functions, but I want to compare the total cost of 6-12 months project including creatives, ads, and commissions to the channels if exists.

Output condition

  • Include fact links after the paragraph or sentence

The question needs fee tables from the platforms and cost figures for creatives and ads from other sources. It tests whether a model finds both and adds them into one total for 6 to 12 months.

Measured · content density

Which model researched further?

20 websitesHaiku 5.5 at Extra, the widest search
18 to 19Qwen3.8-Max, GPT-6.1 Sol at Max and Grok 4.7
10 websitesHaiku 5.5 at Medium, the narrowest search. At Extra the same model cited twice as many.
One dot per answer. Across: answer length in words. Up: distinct websites cited. A dot high on the chart drew on more sources. The shape is the vendor. Hover or tab to a dot for its numbers.
Show as a table
AnswerWordsWebsites 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.

Not measured yet · cited links

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.

145cited links across 8 answers, none opened yet
0answers with no links in the saved copy
0passages graded

Cited links · one square per link, none opened yet

One hollow square per cited link, in the order of the answers table. The squares count links, not websites: an answer that cites one page ten times has ten squares and one website. When the check runs, each square turns grey, red, or red-outlined, and this chart becomes the one a checked comparison shows.

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?.

Our reading · Dorado's view, not a measurement

What this means when you pick a model for research.

  1. Haiku 5.5 doubled its websites with a small gain in words.

    At Medium it cited 10 websites in 1,036 words. At Extra it cited 20 in 1,158 words.

  2. Six answers cite Amazon's seller pages, and five cite Rakuten's.

    Haiku 5.5 at Medium cites neither, and no Yahoo page either. Commission rates sit on the platforms' own pages, so those are the first pages to open.

  3. The shortest answer did not search the least.

    GPT-6 Luna at Max wrote 1,007 words and cited 14 websites. Haiku 5.5 at Medium wrote 1,036 words and cited 10.

All 8 answers

Every answer, with its numbers.

ModelSettingAskedWordsWebsitesLinks
GPT-6 LunaOpenAIMax8 Oct 20261,0071415
GPT-6.1 SolOpenAIMax8 Oct 20261,6591825
Gemini 3.6GoogleThinking8 Oct 20261,4971213
Grok 4.7xAIFast8 Oct 20261,3561819
Qwen3.8-MaxAlibabaThinking8 Oct 20261,1151921
Haiku 5.5AnthropicMedium8 Oct 20261,0361015
Haiku 5.5AnthropicExtra8 Oct 20261,1582023
GPT-5.6 SolOpenAIMax8 Oct 20261,4991414

Haiku 5.5 was asked at two settings. Grok 4.7 has an earlier saved copy that the library replaced with a corrected one at the same address, and the table shows the corrected copy. Each model name opens that answer in full. Each model name opens that answer in full.

The research plan
$50 first month, then $200 a month

Want the sources checked for your product?

These answers are unchecked. In the research plan, we open the sources, check the dates, and correct what is wrong before the brief reaches you.