Chinese AI startup DeepSeek has resumed access to the API after halting its services for nearly three weeks due to capacity constraints.
On Tuesday, the company allowed customers to replenish credits they use in the API. This allows developers to build apps and services on top of the cloud-hosted version of Deepseek’s AI. Server resources remain tense during the day, but company representatives warned in a WeChat message seen by Bloomberg.
Deepseek became famous earlier this year after the release of the openly available R1 “Reasoning” model, which matches or follows the best performance of some of OpenAI’s top models. Deepseek’s competitiveness has encouraged Openai to consider more open-sourcing technologies and “pull up” certain product releases.
As Bloomberg points out, Deepseek’s domestic rivals are also stepping up model production. On the same day Deepseek resumed top-ups of its APIs, Chinese tech giant Alibaba began previewing its latest inference AI model QWQ-Max, which it plans to open source.