Breaking News: Ox Alpha Mystery Solved
On August 27, 2026, the mysterious AI model "Ox Alpha" that captivated developers over the weekend finally revealed its identity—it belongs to Chinese company Z AI, officially named "GLM-5.3-Flash." This model shocked the market with open weights and extremely low pricing, charging only one-tenth of comparable competitors.
More remarkably, Z AI announced that its record-breaking usage was entirely powered by domestically manufactured chips. This news dispelled concerns about China's AI industry being constrained by hardware limitations, proving that local chips can already support large-scale AI services.
Technical Highlights: The King of Cost-Effectiveness
GLM-5.3-Flash quickly topped the OpenRouter platform, with usage double that of the runner-up. Independent analysis shows that while maintaining high intelligence levels, the model offers exceptional cost-effectiveness:
- Price Advantage: Charges only 10% per million tokens compared to competitors
- Performance: Achieves 63% on the DeepSWE benchmark, comparable to Fable 5, but consumes fewer tokens per task
- Context Window: Supports 1 million tokens, suitable for long-horizon agent tasks and complex reasoning
- Multimodal Support: Can process text, images, and other input types simultaneously
Early evaluations showed the model achieving 80% on a specific subset of DeepSWE. Although the comprehensive test result was 63%, considering its extremely low cost and the fact that it runs entirely on domestic chips, this performance is already impressive.
PLUS Tutorial: Setting Up Automated Workflows
The low cost of GLM-5.3-Flash makes it ideal for building automated workflows. Here's a four-step practical method:
Step 1: Create Local Directory and Project
Create a dedicated folder on your computer to store files that need automation. Then launch a new project in the desktop application, selecting this folder as the primary data source.
Step 2: Describe Your Objectives
Clearly explain to the AI system what you want to achieve. For example: "Convert meeting transcripts into actionable task lists, sorted by priority." The system will propose structural suggestions based on your description.
Step 3: Upload Initial Files for Testing
Upload 1-2 sample files and run a single processing flow. Check if the output meets expectations, and adjust prompts or parameters if needed.
Step 4: Create Scheduled Tasks for Continuous Automation
Once satisfied with the single-run results, instruct the system to generate scheduled tasks. You can set it to process new files hourly, daily, or weekly. A useful tip is to convert these scheduled tasks into callable skills using specific command shortcuts.
Pro Tip: For small teams, this type of automated workflow delivers even greater value. Large companies often move slowly, while small teams can quickly test and implement solutions. The focus should be on "buying back time"—automating mundane tasks so employees can focus on high-value work. For employees resistant to new technology, deploy background automations that handle tedious chores without requiring them to learn new interfaces.
Perspective: The Breakthrough Significance of Domestic Chips
The success of GLM-5.3-Flash is not just a product victory, but an important milestone in the self-sufficiency of China's AI industry chain. Previously, there were widespread concerns that US export restrictions on high-end chips would hinder China's AI development. But Z AI has proven that even using entirely domestic chips, it can provide AI services competitive on the international stage.
What does this mean for the global AI market?
- Price wars may come: If domestic chips can reduce costs to this level, other vendors will be forced to follow with price cuts
- Technology diversification: No longer relying on a single supply chain helps industry health
- Local deployment opportunities: Low costs make advanced AI capabilities affordable for SMEs
For developers, now is an excellent time to learn. The open weights of GLM-5.3-Flash mean you can study its architecture in depth, or even run it on your own device. The unified memory architecture and efficient token utilization also provide new ideas for optimizing local deployment.
In the coming months, we may see more applications built on GLM-5.3-Flash emerge. Whether for code assistants, content generation, or data analysis, this model could become the new default choice.