Fetching from the wire…
Public story · 2026-08-03 · high
Pricing runs on four model tiers instead of per seat, and the platform slots into DingTalk's 20 million business accounts.
Why now: The beta landed August 3, folding three already-shipped Alibaba products into one ahead of its rollout inside DingTalk's app for 20 million businesses.
Alibaba launched QwenWork in public beta on August 3, per its Alizila newsroom. The platform merges three existing Alibaba agent tools, Qoderwork, Mulerun and Wukong, into one workplace hub.
The platform is also headed into DingTalk, the messaging app Alibaba says serves more than 20 million enterprises. That's real distribution, skipping the pitch to IT buyers by flipping on one hub inside an app millions of businesses already have open.
QwenWork ships two ways: through a browser or a desktop client with direct access to the local machine.
Both come as two subscription editions layered with a credit system, with four selectable model tiers: Economy, Basic, Advanced and Flagship. A buyer pays more for a task routed to a heavier model, not for adding another employee seat. Alizila's announcement doesn't list actual credit costs or per-tier pricing, so how much a Flagship task costs versus an Economy one is still unknown.
A desktop client with standing access to the local machine is one more agent identity to track. Related coverage on agent-bound identity across MCP, LLM and API calls raises the same broader problem, though it doesn't address desktop clients specifically.
Charging by model tier instead of by seat is Alibaba admitting the seat is the wrong unit for agent software. If the Economy-to-Flagship menu sticks and buyers accept paying per task complexity, expect Microsoft and Google to copy the same lever within a year.
Each link below shares sources, entities, or timing with this story.
Same source
Cite the same source (Alizila (Alibaba newsroom)).
Semantically similar
Covers closely related ground (similarity 0.77).
Covers closely related ground (similarity 0.71).
Covers closely related ground (similarity 0.70).
Covers closely related ground (similarity 0.69).
Covers closely related ground (similarity 0.69).
Covers closely related ground (similarity 0.68).
Covers closely related ground (similarity 0.68).