Odaily Planet Daily reports: Zhipu has officially launched and open-sourced GLM-5.3-Flash, the first native multimodal model in the GLM-5 series.
According to reports, the overall performance of GLM-5.3-Flash exceeds that of GLM-5.2, with programming and agent evaluation scores approaching those of Claude Opus 4.8, while costing only one-tenth of GLM-5.2. It also features a new foundational model, introducing for the first time in the GLM main series a hybrid architecture combining sparse and linear attention, and has been pre-trained on 30T tokens of multimodal data.
Zhipu stated that, to gather extensive and professional feedback from users, it conducted large-scale testing of the anonymous model Ox-Alpha (known in the Chinese community as Niu Lai) on OpenCode and OpenRouter prior to its official release. Ox-Alpha quickly became the most popular model of the week, setting new records for usage volume on both platforms. All of this request traffic was powered by domestic chips.
Previously, the community reported that Ox Alpha’s DeepSWE achieved 80% on a small-sample test consisting of only 10 questions. After expanding the sample size, the score dropped to approximately 63%, and the testers themselves corrected their initial claims. This result still places it in the top tier, though not as impressive as the original 80%. With the weights now open-sourced, developers can directly deploy the model using frameworks such as vLLM, SGLang, and KTransformers, eliminating the need to rely on anonymous model APIs.