Quick Verdict
QwenWork is Alibaba’s entry into the AI office agent race, opened to public beta in August 2026. Its official positioning is blunt: not just conversation, but delivery. You describe a task in one sentence and it hands back a usable PPTX, Word doc, Excel sheet or HTML page — not a wall of text you still have to assemble yourself.
There is really only one precondition worth checking first: is your team already on DingTalk? QwenWork can be invoked directly inside the DingTalk workspace with no separate login or redirect, and that entry point is something its rivals simply cannot offer. If your organization runs on Feishu instead, ByteDance’s Doubao Work will fit better; if you are in the Tencent ecosystem, look at WorkBuddy.
Three things deserve scrutiny before you commit. First, it bills by credits, not by seats or usage time, and heavy users can burn through a month’s allowance well before the month ends. Second, it only opened in August 2026 and two headline features are not evenly available yet — full-stack web building works on web and DingTalk but not the desktop client, while browser automation is the exact opposite and works only in the client. Third, it is a different product from Qwen: that one is a chat assistant, this one is an agent that does work and delivers files. Do not buy the wrong one.
Best For
- Teams already running on DingTalk — in-workspace invocation is the one advantage nobody else can copy
- Roles that need deliverables, not answers — finance, HR, marketing, operations and engineering staff who have to produce reports and decks
- Business analysts who want data aggregation, analysis, charting and publishing handled end to end
- Heavy individual users — Alibaba explicitly targets students writing papers, job seekers polishing résumés and freelancers building proposals
- Poor fit for: people who only need Q&A (free Qwen is enough); organizations locked into Feishu or WeCom; teams that cannot tolerate an early-stage product
Key Features
- End-to-end delivery — one instruction to a finished artifact: documents, spreadsheets, decks, code
- Flagship model backing — built on the Qwen model family, with million-token context and native multimodal input
- Deep DingTalk integration — invoked inside the DingTalk workspace, with 25 enterprise IM capabilities claimed
- Local file read/write from the desktop client — reaches documents, spreadsheets, images and video on your own disk, supports drag-and-drop and right-click menus, and saves results back to the original path
- Full-stack web building — generates interactive pages and connects a database on demand (web and DingTalk only for now)
- Scheduled cloud tasks — runs on a timer in the cloud and pushes results to DingTalk
- Browser automation — clicks, types and scrapes on web pages (desktop client only)
- Skills and memory — save a working flow as a reusable skill, backed by personal cloud storage and connectors
Use Cases
- Decks from a sentence — give a topic and key points, get an editable PPTX rather than an outline
- Batch local file work — right-click a folder of spreadsheets, have them cleaned, merged and analyzed, results written back in place
- Data dashboards — turn business data into an interactive page, add a database and a password when needed
- In-DingTalk productivity — assign work from the workspace and forward the output straight to colleagues
- Recurring reports — schedule a weekly cloud task that pulls data, writes the report and pushes it to DingTalk
Pricing
Figures below reflect the official pricing page as of 2026-08-27. Listed prices are the auto-renewing monthly rate; monthly-without-renewal and annual (30% off) options also exist.
| Plan | Auto-renew monthly | Monthly credits | Daily login bonus | Storage | Publishable pages |
|---|---|---|---|---|---|
| Free | ¥0 | none | 100/day | 1GB | 5 |
| Plus | ¥78 (list ¥98) | 2,000 | 200/day | 10GB | 10 |
| Pro | ¥158 (list ¥198) | 4,000 | 300/day | 20GB | 50 |
The gap is wider than credit counts suggest: the free tier cannot buy credit packs, does not support access passwords or database storage, and caps queued parallel tasks at 10, while paid tiers remove that cap. All new accounts get 2,000 signup credits valid for 90 days, flagged as a limited-time promotion.
An enterprise tier exists behind its own signup flow, but no pricing or seat allowances are published — you have to request a quote. Credit pack pricing is likewise unlisted.
Practical advice: run your single most typical task on the free tier and watch how far 100 daily credits actually goes. That number decides whether to pay, and which tier, better than any spec sheet.
Pros
- Ships usable files rather than text you have to rework
- Zero-redirect invocation inside DingTalk is a genuine structural advantage
- Desktop client reads and writes local disk and saves back in place, removing upload/download churn
- Free tier includes daily credits, enough to run real tasks rather than a demo
- Published pages can connect a database and take an access password
Cons
- Credit-based billing with opaque consumption rates makes monthly cost hard to forecast
- Capabilities are unevenly split across web, desktop and DingTalk clients
- Neither enterprise pricing nor credit pack pricing is public
- Public beta only since August 2026; stability and completeness are still early
- Its biggest advantage is tied to DingTalk, leaving non-DingTalk teams with far less reason to choose it
Alternatives
| Tool | Better for | Strengths | Weaknesses |
|---|---|---|---|
| Doubao Work | Feishu organizations | Deep Feishu integration, local and cloud PC modes | Launched even more recently |
| WorkBuddy | Tencent ecosystem users | Longer on the market, real user scale | Ecosystem coupling cuts both ways |
| Qwen | Q&A and writing only | Free, no learning curve | Does not deliver files or execute tasks |
| Teable | Teams building data apps | Open source, self-hostable, fully private | Not a general office agent |
FAQ
Is QwenWork the same thing as Qwen?
No. Qwen is a chat assistant that answers questions and writes copy. QwenWork is an office agent that decomposes tasks, calls tools and delivers finished decks, spreadsheets and web pages. They share the Qwen model family but are separate products with separate pricing.
Is the free tier enough?
Enough to evaluate, not enough for daily work. There are no monthly base credits — only 100 daily login credits that expire the same day and do not accumulate, plus a 10-task parallel cap and no credit pack purchases. Use it to run your highest-frequency task, then decide.
How fast do credits burn?
Alibaba publishes tier allowances but not per-task consumption rates, so budgeting has to be empirical. Run a few real tasks on the free tier, record the credit delta, and work backwards to the tier you need.
Do I have to install the desktop client?
For full capability, yes. Browser automation is client-only, while full-stack web building is currently web and DingTalk only. The web app works without installation but gives up local file access and automation.
Is it worth it if we do not use DingTalk?
You can use it, but understand what you are giving up. Without DingTalk it becomes a general file-delivering agent, and at that point it should be compared head to head with Doubao Work and WorkBuddy rather than chosen by default.
What does the enterprise plan cost?
Unpublished. Enterprise access has its own registration flow and requires a quote. The ¥78/¥158 personal rates do not apply to enterprise seats, so confirm separately before budgeting.
Bottom Line
QwenWork is Alibaba’s answer in the office agent race, and the product thesis is clear: push AI from answering questions to shipping finished work. Local file read/write and database-backed page publishing are pragmatic, well-judged choices.
But it has only been in public beta since August 2026, capabilities are not yet even across its three clients, and there is no public reference point for credit consumption. The sensible move right now: if you are inside DingTalk, try the free tier immediately; if you are not, do not rush to pay. All three major Chinese vendors are iterating fast here, and waiting a month or two costs far less than picking wrong and migrating later.