Ox Alpha steps out of the shadows as Z.ai’s GLM-5.3-Flash

Chinese AI company Z.ai has confirmed that the previously anonymous Ox Alpha model is GLM-5.3-Flash, a new multimodal member of its GLM model family. The company says it plans to release the model’s weights, allowing developers to run, adapt and build products with the system.

Luz Ding reports for Bloomberg that Ox Alpha rose rapidly to the top of OpenRouter’s usage leaderboard after appearing without an identified creator. The model more than doubled DeepSeek’s usage on the marketplace, according to the report. Z.ai offered it free for one week and has not announced pricing after that period.

In an official blog post, Z.ai says it also tested GLM-5.3-Flash anonymously as Ox Alpha on OpenRouter and OpenCode to collect feedback. The company describes the model as built for coding, AI agents and work involving text, images and video.

Designed to reduce inference costs

Z.ai says GLM-5.3-Flash has 320 billion total parameters, but activates 18 billion at a time. That design aims to lower the computing resources needed for each request. The company claims the model outperforms its earlier GLM-5.2 model at one tenth of the price.

The company also says the model approaches Anthropic’s Claude Opus 4.8 on some coding and agent benchmarks. These are vendor-reported results and should not be treated as independent comparisons.

GLM-5.3-Flash uses a hybrid attention architecture intended to handle long documents and other large inputs more efficiently. Z.ai says it supports context windows of up to one million tokens. It adds that it served the Ox Alpha traffic using Chinese AI chips and achieved costs comparable to mainstream Nvidia GPU systems after optimizing its infrastructure.

Sources

• Z.ai confirms Ox Alpha is a new iteration of its GLM series and says it will release the weights for it tonight; Ox Alpha topped OpenRouter’s leaderboard – Bloomberg
• Z.ai releases GLM-5.3-Flash, the first natively multimodal GLM-5 series model, with 320B parameters, saying it outperforms GLM-5.2 at “one-tenth the price” – Z.ai