We put the questions a buyer in Taiwan might ask to free AI models that OpenRouter assigns at random. The index tracks whether brands appear and where, giving brand teams an observation they can inspect.
Edition 2026-Q4
Collection started (UTC)
Collection completed (UTC)
120 successful answers from 11 different models
Plan: 5 industries × 3 questions per industry × 8 runs per question, or 120 calls. OpenRouter's free router picks the model for each call at random.
Five industries, one comparison method
“Taiwan brands” means brands serving the Taiwan market, including international brands. These lists define the scope of this edition; they do not cover every option in the market.
E-commerce platforms
24 successful answers
Sorted by mention rate, then average position. Positions count only brands tracked in this edition.
OpenRouter Free Models Router · Question 3 · Run 3
Requested at (UTC): Model returned by the API: nvidia/nemotron-3.5-lightning:free
1. Shopee Taiwan – Frequent free-shipping promotions and Shopee Mall’s buyer-protection coverage make daily-item delivery affordable and returns relatively straightforward within the app’s protection period.
2. PChome 24h Mall – Nationwide integration with convenience-store pickup/return networks enables 24‑hour delivery and easy in‑person returns, making it highly practical for everyday purchases.
3. Momo Online – Backed by Sinyi department‑store locations, allowing direct in‑store returns and quality‑checked daily products, though free‑shipping thresholds are often higher for small orders.
OpenRouter Free Models Router · Question 3 · Run 4
Requested at (UTC): Model returned by the API: apodex/apodex-1.1-mini:free
OpenRouter Free Models Router · Question 3 · Run 6
Requested at (UTC): Model returned by the API: nvidia/nemotron-3.5-lightning:free
由於台灣各家數位銀行的「免費跨行轉帳次數」會隨帳戶等級、促銷活動、個月限制以及交易類型（如櫃位轉帳 vs. 應用程式轉帳）而頻繁變動，且我的資料無法即時對應最新的條款與促銷細節，因此我無法基於目前的可驗證資訊，可靠地列出具體品牌並附上推薦理由。
依照「若資訊不足，請明確說明」的原則，我在此明確說明： **目前無法提供經核實的免費轉帳次數與品牌推薦清單**。建議您直接查詢各家銀行最新的費用說明（如開線銀行、富邦數位銀行、PChome銀行等），或依照自己的轉帳頻率與需求（如每月固定次數、特定時間段內轉帳等）再比對最新的免費額度與費率。若您有具體的帳戶類型或轉帳習慣，我可以協助說明一般比較方式或幫您快速查詢特定銀行的最新政策。
OpenRouter Free Models Router · Question 3 · Run 7
Requested at (UTC): Model returned by the API: google/gemma-4-26b-a4b-it:free
針對每月頻繁跨行轉帳的需求，以下是根據目前台灣市場優惠條件與轉帳便利性，依推薦程度排序的數位帳戶建議：
1. **台新 Richart Digital Account**：轉帳優惠最透明且容易達成，常有針對特定條件（如定期存款或薪轉）提供大量免費轉帳次數的活動。
2. **國泰世華 Cube Digital Account**：App 介面體驗極佳且功能直覺，轉帳流程順暢，對於追求數位操作便利性的用戶非常友善。
3. **聯邦 Bank Bankee Digital Account**：轉帳手續費優惠門檻相對親民，是適合小額、高頻率轉帳且不希望被複雜條件綁架的穩健選擇。
4. **永豐 Bankee Digital Account**：整合數位支付與轉帳功能非常完善，對於經常使用行動支付與跨行金流整合的用戶來說極具效率。
**提醒：** 數位帳戶的免費轉帳次數與條件（如需滿足薪轉、繳費或存款額度）會隨銀行政策變動，建議在開戶前於各行官網確認最新的「優惠活動公告」。
OpenRouter Free Models Router · Question 3 · Run 8
Requested at (UTC): Model returned by the API: nvidia/nemotron-3.5-lightning:free
OpenRouter Free Models Router · Question 1 · Run 6
Requested at (UTC): Model returned by the API: nvidia/nemotron-3.5-lightning:free
1. Agoda – Extensive Japan hotel inventory with frequent flash sales and seamless English/Traditional Chinese interface tailored for Taiwan travelers.
2. Booking.com – Global reliability with detailed filters, guest reviews, and flexible cancellation policies for Japanese city centers and ryokans.
3. Klook – Integrated platform for hotels + local activities/transfers, offering curated Japan experiences and bundle discounts for Taiwan outbound travelers.
4. Trip.com – One-stop booking for flights, hotels, and JR Pass alternatives, often featuring competitive bundle pricing and Chinese-language support for Taiwan users.
5. Rakuten Travel – Japan-domestic hotel network with points system and ryokan selections, useful for regional stays beyond major tourist hubs.
OpenRouter Free Models Router · Question 1 · Run 7
Requested at (UTC): Model returned by the API: nvidia/nemotron-3-ultra-550b-a55b:free
Start with an AI visibility check of your website, then audit the answers to questions your buyers ask. Tenten’s engineers and marketers work inside your workflow to get a specific improvement live.
Every model receives the same Traditional Chinese question and the suffix below. Each call starts a new, single-turn conversation with no earlier answers. Temperature is omitted, leaving the provider default; max_tokens is 4000 (reasoning models count their reasoning tokens toward this limit). Calls time out after 60 seconds, with up to two retries for HTTP 429 or server errors and a maximum of four concurrent calls. Failed calls are retained separately.
We call OpenRouter's free models router (openrouter/free), which picks one of the currently available free models at random for each call. The index therefore reflects answers from a mix of free models, not from ChatGPT, Claude, Gemini or Perplexity, and the mix can change between editions. None of these calls use web search. The router sometimes picks a content-safety classifier that only returns a verdict such as “User Safety: safe”; those replies are not counted as answers and the call is repeated.
OpenRouter Free Models Router · No additional search tools enabled
We normalize text with NFKC, including full-width characters, and lowercase Latin letters before matching names and public aliases. Aliases containing CJK characters match as substrings; Latin aliases need word boundaries, so Applebee’s does not count as Apple. Longer aliases win overlaps, and each brand counts once per answer. Product lines and bank names can also be aliases; each industry discloses its full alias list.
How rates and positions are calculated
Mention rate is the share of successful industry answers mentioning a brand. Average position is the mean order of first appearance among tracked brands, only when mentioned; it is not the answer’s original list number. First-place share is the share of successful answers in which the brand appears first. Remaining ties are sorted by brand name.
What these results can tell you
Each industry has a small sample. Prompt selection, collection time and which model the router picks for each call all affect results, so this is not a stable view of the entire market.
API answers can differ from consumer apps with personalization, location, subscription plans or web search. This edition's answers come from randomly assigned free models and do not represent any specific AI assistant.
Alias matching can miss nicknames and count comparisons, negative mentions or other non-recommendation contexts. It is not a human review of sentiment or factual accuracy.
We retain at most 4,000 characters per answer and analyze that text. Token limits and character truncation can omit brands later in an answer.
Rankings describe brand mentions in these AI answers. They do not measure market share, product quality or service ratings, and are not buying advice.
Independence and reuse
No brand paid to participate or was contacted for this index. Brand lists and prompts are fixed before collection and are not changed to secure a brand a place in the rankings.
The index data is available under CC BY 4.0 for sharing, adaptation and commercial use. Credit “Tenten AI Taiwan Brand AI Visibility Index,” include the edition and this page’s URL, link to the license and identify any changes. Keep the incomplete status when reusing unfinished data.
No. This edition tracks brands serving Taiwan, including local and international brands. Questions are in Traditional Chinese and use Taiwan consumer and buyer scenarios.
Does a higher mention rate mean a better brand?
Mention rate is only the share of these successful answers mentioning a brand. It does not measure market share or quality. Comparisons and negative mentions can also match, so these results are not buying advice.
Why do I get different answers in ChatGPT?
This index does not test ChatGPT. Each question goes to a free model that OpenRouter picks at random. ChatGPT and other consumer apps use different models, personalization, location, search tools and conversation history, so their answers differ.
When are rankings unavailable?
Industries without successful answers do not show rankings. While collection is underway or only some calls succeed, results show the current sample size and an incomplete status. A complete edition requires the full collection.
Can I cite or share the index?
Yes. The data is licensed under CC BY 4.0. Credit Tenten AI, include the edition, page and license links, and identify any changes. When citing results, say that the answers came from randomly assigned free models, and keep the prompts and collection dates as context.
Bring the result to a conversation. We will tell you which issues matter most, which ones you can fix yourselves, and which ones are worth doing together.