Alibaba did the thing you don't expect from a company with a flagship model: it put that flagship on the same shelf as everyone else's and took down the 'ours only' sign.
QwenWork, Alibaba's office AI suite, opened a choice of ten models on November 16. The list runs Qwen4-Max, Kimi-K3.1, GLM-5.3, DeepSeek, MiniMax, Step. Before that, users got three fixed tiers: standard, advanced and Qwen3.8-Max. Now it's a menu.
The logic is blunt. Models handle tasks differently and cost differently. Long multi-step reasoning — take Qwen4-Max. Rule-checking, plan review, 3D layouts — take GLM. Cheap tokens for routine work — take whatever is cheapest. Nobody is selling you 'the best model'. They're selling a choice with a price tag.
QwenWork isn't a startup with nothing to lose. It's an Alibaba product, and Qwen is Alibaba's own model, its flagship, its shop window. Letting Moonshot's Kimi, Zhipu's GLM and DeepSeek onto that window is an admission: the model stopped being the moat. The moat is the product, the access, the habit.
For enterprise buyers there's an admin layer too: model access is granted by department, group or individual member. IT no longer picks a vendor. It picks a model per team. Procurement starts to look like a menu.
The West has tried pieces of this — model routers, aggregators, bring-your-own-key. But those were small outfits sitting on top of someone else's API. Here the model owner itself stacks rivals next to its flagship. Different signal: the bet is on the interface and the workflow, not on the weights.
One 'choose your model' button hits the price list harder than any press release. When quality and cost per token sit side by side, people compare. Being best isn't enough — you have to be best at this task and not much pricier than the next item in the dropdown.
Which leads to the uncomfortable conclusion: office AI is drifting toward the model as a consumable. Not 'our model is smarter' but 'our shell picks smarter'.
Western office suites mostly let you choose within their own line. Alibaba opened the shelf to outsiders. Different bet. Retention, not benchmarks, will settle it.
There are no revenue or user numbers in the reports. So the scale is still unknown. The direction isn't.
1. Does Alibaba lose Qwen's differentiation, or buy loyalty to the shell instead? What would have to happen to flip your answer?
2. If models get picked by price per token, who blinks first — Kimi, GLM or DeepSeek? And does choice become an illusion once everyone just accepts the default?
3. Per-department admin: a gift to IT, or a quiet way to keep staff off 'the wrong' models?
Prediction: Within a year, at least one Western office suite will open a picker for third-party models — otherwise the price comparison goes against it.
Disclaimer: no official Alibaba announcement in the cluster; compiled from Pandaily and Leiphone — not officially confirmed.
Source: pandaily.com · post in Telegram