What is the cost advantage of GLM-5.3-Flash?

The cost per task is about 7.5 times lower than the top model GLM-5.3, according to Artificial Analysis.

How does GLM-5.3-Flash achieve long-context efficiency?

It combines linear attention and sparse attention to compress past context and focus on key tokens, balancing accuracy and cost.

Large Language ModelsAI Business & IndustryITmedia AI＋Aug 27, 2026Aug 27, 2026, 16:00 JST

Z.ai reveals 'Ox Alpha' is GLM-5.3-Flash, runs on China-made chips

Send on LINE

3 Key Points

What happened
Z.ai revealed on August 26 that the anonymous model "Ox Alpha" was actually its own GLM-5.3-Flash, and that inference during the validation period ran on a cluster of China-made AI chips. The model's weights were released under a commercial-use MIT license the same day.
Why it matters
GLM-5.3-Flash accumulated 17.5 trillion tokens on OpenRouter since August 20, ranking #1 in the weekly chart. Z.ai says the cluster of tens of thousands of China-made chips handled this demand at a per-token cost on par with Nvidia GPUs, arguing it shows China-made chips can support state-of-the-art inference at scale.
What to watch
The model has 320B total parameters and 18B active, supports text, image, and video input, and up to 1M token context. It scored 57 on the Artificial Analysis Intelligence Index, just 3 points behind the top model GLM-5.3, with about 7.5x lower cost per task.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

Z.ai has been offering GLM-5.3-Flash anonymously on OpenRouter since August 20, sparking speculation about its origin. The reveal on August 26 that it is a Z.ai model running on a cluster of tens of thousands of China-made chips is significant because it demonstrates that domestic hardware can serve frontier-level inference at a cost comparable to Nvidia GPUs. The model's performance—scoring 57 on the Artificial Analysis Intelligence Index, just 3 points behind the top model GLM-5.3—and its lower cost per task suggest that the combination of efficient architecture and China-made chips could offer a cost-effective alternative in the AI inference market. This development may encourage further investment in domestic chip ecosystems, though the article does not speculate on broader market impacts.

FAQ

What is the cost advantage of GLM-5.3-Flash?: The cost per task is about 7.5 times lower than the top model GLM-5.3, according to Artificial Analysis.
How does GLM-5.3-Flash achieve long-context efficiency?: It combines linear attention and sparse attention to compress past context and focus on key tokens, balancing accuracy and cost.

ITmedia AI＋Read Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.