On 27 August 2026, Zhipu AI (Z.ai) officially announced the GLM-5.3-Flash model. The open-source language model is meant to excel above all at programming tasks and, according to the vendor, can compete with established models such as Claude.
Technical specifications
GLM-5.3-Flash uses a Mixture-of-Experts architecture with 320B-A18B parameters and a context window of 1,048,576 tokens. At more than 300 GB the model is correspondingly large, supports native visual capabilities and a hybrid combination of linear and sparse attention. It is released under the permissive MIT license. During its anonymous test phase in early August 2026 it ran under the name “Ox Alpha”.
Benchmark claims with caution
According to the vendor, GLM-5.3-Flash solves coding tasks on the first attempt in around 80 percent of cases; Claude is cited at 65 percent across four attempts. In the Artificial Analysis Intelligence Index the model reportedly scores 57, putting it on a level with Claude Opus 4.8. These figures come from the vendor and the index respectively and have not yet been broadly confirmed independently.
Pricing and market
Via the international API, GLM-5.3-Flash costs 0.15 US dollars per million input tokens and 0.50 US dollars per million output tokens; coding plans start at 18 US dollars per month. A promotional discount runs until 9 September. Following the announcement, Zhipu’s stock rose by 12.62 percent.
Sources: Borns IT- und Windows-Blog.
