In this issue
The short version
- Claude Sheets and Whiteboards are in beta and Claude Data is generally available. Sheets has real limits, including no pivot tables, no macros and a 100,000-cell cap.
- Tencent's Yuanbao Search now has a consumer site at so.qq.com that reaches WeChat articles and Channels videos as well as the open web.
- Anthropic's IPO roadshow sets a preliminary range of $70 to $78 a share. For tool buyers, the useful part is the disclosure that comes with it.
- AI token bundles sold well on Singles' Day. Buy them from the model provider or an authorised reseller, never from an unlicensed API relay.
Four stories from November 12 point at the same question: where should everyday AI work actually happen? Claude moves into spreadsheets and whiteboards, Tencent (腾讯) turns its assistant's search into a standalone product, Anthropic opens its finances to public investors, and Chinese shoppers bought AI tokens alongside laptops. Each section explains what changed and which tools in the directory are worth comparing.
01
Claude Sheets, Whiteboards and Data: Anthropic moves into office documents
What Anthropic released
Anthropic began rolling out three workspace tools inside Claude on November 10. Claude Sheets builds an editable spreadsheet from a prompt, with tabs, formulas, formatting and dropdowns. Claude Whiteboards is a shared board where a team adds notes and Claude groups related ideas into themes and a plan. Claude Data, previously called Claude Dashboards, answers plain-language questions about connected company data with charts and short narrated summaries.
Sheets and Whiteboards launched in beta; Data is generally available on paid plans. On Enterprise plans, Sheets and Whiteboards start switched off, so an administrator decides when teams can use them.
The limits matter as much as the features. Anthropic's Sheets documentation lists no pivot tables or Excel macros and a 100,000-cell cap. Sheets also cannot yet pull live data from connected apps, and the help pages describe mobile access to a sheet as read-only. Claude Data reaches its full value only when a data connector such as BigQuery or Redshift is in place. Press coverage framed the launch as a challenge to Microsoft Office. For now, these tools are better understood as quick ways to draft a tracker or a first analysis that you then check.
Choosing a spreadsheet assistant for the job
Start from the job, not the brand. If you need a working tracker quickly and already use Claude, Sheets removes the step of copying a table out of a chat. If your workbook already lives in Excel or Google Sheets and you only need help with formulas, a focused tool is lighter: Excelly-AI and SheetGod turn a plain-English request into a formula you paste into your own file. ChatExcel edits and analyses an uploaded spreadsheet through chat. More options are in the Data Analysis category.
Test any of them with the same small brief. Take a non-confidential table of 50 rows, such as expenses with a date, category and amount, and ask for three things: a monthly total per category, a column flagging any amount above a threshold you set, and one sentence on the biggest change. Then check the output. Recalculate one total by hand, confirm the flag formula refers to the right column, and make sure the summary sentence matches the numbers. A tool that gets the formula right but writes a confident wrong summary still needs human review. Our Excelly-AI review covers the formula-only route in more detail.
Get HAI Daily in your inbox
The day's AI news, what it changes, and which tools to try.
02
Tencent Yuanbao Search: AI search that reaches inside WeChat
What Tencent launched
Tencent opened Yuanbao (元宝) Search to developers on October 20 through an API, MCP and Skill integrations. Tencent describes an index of hundreds of billions of pages and 157 authoritative sources across 13 sectors, including finance, healthcare, law and travel. On November 11, its 28th anniversary, Tencent launched a consumer site at so.qq.com. It searches the open web together with WeChat (微信) Official Account articles, Channels videos and Tencent News content, and adds an academic tab and an AI summary at the top of results.
The distinctive part is access to Tencent's own ecosystem. A large share of Chinese-language commentary, company announcements and how-to content is published as WeChat articles that general web search engines index poorly. One Sina Finance commentary noted that the 157-source partnerships are not exclusive, so the real difference is Tencent's own content.
How to evaluate an AI search tool
AI search is only as useful as its citations. When you compare Tencent Yuanbao with an answer engine such as Perplexity, run the same three queries in each. Use one recent event, one question that needs an official source (a regulation or a product's pricing page), and one niche question in your own field. For every answer, open the cited sources and check that they say what the summary claims, that the dates are current, and that at least one source is primary.
Language and region decide a lot. For Chinese-language research that depends on WeChat content, Yuanbao has an access advantage. For English-language sources, a tool built around the open web may cover more. Browse the Search Engines category for more options, and see our Perplexity vs ChatGPT comparison for the difference between an answer engine and a general chat assistant.
03
Anthropic's IPO roadshow: what the filing means for tool buyers
The offering terms
Anthropic launched the roadshow for its initial public offering on November 10. According to reports of the filing, the company plans to sell 825 million Class A shares at a preliminary range of $70 to $78 each, and has applied to list on the Nasdaq Global Select Market under the ticker ANT. Separately, seven technology companies agreed to buy about $26 billion of shares in a concurrent private placement. The price range is preliminary, and the registration statement had not taken effect when this issue was written, so the final price and listing date may change.
This is not investment commentary. We cover the IPO because a public listing changes what customers can learn about a supplier they depend on.
Use public disclosures in vendor due diligence
When an AI vendor files to go public, its prospectus describes risk factors, customer concentration, spending on compute and how the company is governed. Those details help an IT or procurement team judge supplier risk, alongside the usual checks on data handling and service terms.
If your team relies on Claude or any other assistant for daily work, keep a one-page vendor note. Record which plan you use, where your data is processed, the retention and training settings you chose, the service-level terms, and an alternative you could switch to. Update it when a filing, price change or policy update appears. Company-research tools in the Finance AI category can help collect filings, but read the original document before you rely on any figure. Our ChatGPT vs Claude guide is a starting point for the alternatives column.
04
AI tokens on Singles' Day shelves: buying model credits safely
What the sales data shows
This year's Singles' Day sales in China included AI model credits as an ordinary product category. In JD.com's (京东) end-of-festival report, sales of large-model token bundles rose more than tenfold compared with the previous period, and searches for "Token" rose more than thirteenfold. JD said sales in one livestream on November 2 rose 20 times, while AI laptops bundled with token packs doubled. JD's AI software top-up section carries credits from providers including DeepSeek (深度求索), Zhipu (智谱) GLM and Tencent Cloud (腾讯云). On Tmall (天猫), a livestream sold close to 100,000 Qwen (通义千问) office monthly passes in about half an hour.
These are growth rates from a low base. The platforms did not publish unit sales, so the figures show rising interest rather than the size of the market. Coverage of the festival also warned about grey-market API "relay" shops. These resell access through unofficial accounts and can expose prompts and files, overcharge or cut off service without notice.
Buy credits like a utility, not a bargain
Model credits are worth buying when you already know your workload. Before buying a bundle, estimate a typical month: how many requests, roughly how long each prompt and answer, and whether you need a fast model or a reasoning model. Compare that with the provider's own price list. A cheap bundle for the wrong model is not a saving.
Buy directly from the model provider or an authorised reseller named on the provider's site. For DeepSeek or Qwen Chat, that means checking the provider's official page before paying a third party. Never paste an API key into a service you cannot identify. Avoid relays that ask you to route traffic through their own endpoint, because every prompt and file passes through their servers. Start with a small top-up, watch actual usage for a week, then scale up.
What to try next
Choose where your AI work should live
If you mainly need spreadsheet help, decide between an assistant that builds the whole sheet and a formula tool that works inside your existing file. For research, pick the search tool whose sources cover your language and field, and check its citations. For any vendor you depend on, keep a short due-diligence note and update it when new disclosures appear. When buying credits, size your usage first and buy only from official channels.
Compare options in Data Analysis, Search Engines and Chatbots, then test one real task before committing to a plan.
How we put this issue together
The H.AI Tools Team checks each story against the official sources linked in the text and removes claims we could not confirm. Product links lead to H.AI Tools directory profiles. Unless an issue says otherwise, product notes describe vendor documentation, not our own benchmark tests.