DeepSeek's Newest Model Tops Global Weekly Usage With an 801% Jump

DeepSeek's newest model has taken the top spot worldwide in weekly call volume, with usage up 801% compared with the previous week, according to a report from 36Kr. The report rests on two data points: the model's first-place ranking in the weekly call-volume table it cites, and a roughly eightfold week-over-week increase.

Weekly call volume is a rough proxy for how deeply a model has been adopted by developers and product teams, and a jump of this size means new usage arrived within a very short window rather than accumulating gradually over months.

The growth rate also needs to be read against its base. A week-over-week figure describes change between two adjacent periods; it says nothing about the absolute scale of traffic, and it cannot distinguish new integrations from existing applications simply calling the model more often.

For DeepSeek, the significance is that the new version is being pulled into real workloads rather than only scoring well on benchmarks. Production traffic subjects a model to different tests: inference cost, latency stability and the reliability of the API under sustained load.

It is worth remembering that leadership in call volume tends to be unstable. Developers switch models readily, and competition concentrates on price, latency, context length, tool-calling quality and service uptime, so a single release or price cut can reorder the table within weeks.

What to watch next: whether the ranking methodology behind the report is published in more detail, whether the growth holds over the coming weeks instead of proving to be a one-off spike, and whether rival labs respond with price changes or new versions of their own.

The practical takeaway for teams building on these models is that selection is becoming more dynamic. Rather than locking into one provider for the long term, it is worth running small regression tests against your own tasks and keeping switching costs low enough to act on the next shift.