Key Takeaways
- • Zhipu AI's GLM-5.3-Flash model (codenamed Ox Alpha) achieved a score of 57 on the Artificial Analysis Intelligence index, matching Anthropic's Claude Opus 4.8 and outperforming DeepSeek's V4 Pro (score: 53), signaling competitive parity in large language model performance among leading vendors.
- • Zhipu's stock surged 9% intraday to HK$1,124 following the August 26 official confirmation and launch, reflecting investor enthusiasm for the company's technical achievement and strategic positioning in the AI market.
- • The GLM-5.3-Flash model accumulated over 50 trillion tokens in traffic within five days of anonymous testing on OpenRouter and OpenCode platforms, with current usage exceeding twice that of DeepSeek, indicating strong developer adoption and market demand.
- • All inference workloads for GLM-5.3-Flash are powered entirely by domestic chips, with over 100,000 domestic chip units deployed, achieving cost per token and hardware efficiency metrics comparable to mainstream NVIDIA GPUs—a significant validation of China's domestic semiconductor capabilities in AI infrastructure.
Summary
Zhipu AI officially unveiled its GLM-5.3-Flash model (codenamed Ox Alpha) on August 26, 2026, marking a milestone in competitive AI model performance. The model scored 57 on the Artificial Analysis Intelligence composite index, placing it on equal footing with Anthropic's Claude Opus 4.8. This achievement represents Zhipu's strongest benchmark result to date and positions the company ahead of DeepSeek's V4 Pro, which scored 53 on the same index.
Prior to its official release, GLM-5.3-Flash was available for anonymous testing on major AI platforms OpenRouter and OpenCode, where it generated significant traction in the developer community under the informal name "Niu Lai" (Ox Alpha). Within five days, the model processed over 50 trillion tokens—representing a new traffic record for both platforms. Current usage levels have exceeded twice the volume of DeepSeek, indicating substantial developer interest and market adoption momentum.
The market response proved immediate and tangible. Zhipu's Hong Kong-listed stock rose more than 9% intraday to HK$1,124 following the announcement, reflecting investor confidence in both the technical achievement and the company's competitive positioning. Market analysts noted particular interest in the domestic semiconductor angle, which has become increasingly central to Zhipu's strategic narrative.
The most strategically significant aspect of this launch is Zhipu's full reliance on domestic chips for all inference operations. The company deployed over 100,000 domestic chip units to handle the massive inference workload, achieving hardware efficiency and cost-per-token metrics that the company claims are comparable to mainstream NVIDIA GPUs. This development carries substantial implications for China's AI infrastructure independence and represents a validation of domestic semiconductor viability for large-scale commercial AI deployments.
Prior to its official release, GLM-5.3-Flash was available for anonymous testing on major AI platforms OpenRouter and OpenCode, where it generated significant traction in the developer community under the informal name "Niu Lai" (Ox Alpha). Within five days, the model processed over 50 trillion tokens—representing a new traffic record for both platforms. Current usage levels have exceeded twice the volume of DeepSeek, indicating substantial developer interest and market adoption momentum.
The market response proved immediate and tangible. Zhipu's Hong Kong-listed stock rose more than 9% intraday to HK$1,124 following the announcement, reflecting investor confidence in both the technical achievement and the company's competitive positioning. Market analysts noted particular interest in the domestic semiconductor angle, which has become increasingly central to Zhipu's strategic narrative.
The most strategically significant aspect of this launch is Zhipu's full reliance on domestic chips for all inference operations. The company deployed over 100,000 domestic chip units to handle the massive inference workload, achieving hardware efficiency and cost-per-token metrics that the company claims are comparable to mainstream NVIDIA GPUs. This development carries substantial implications for China's AI infrastructure independence and represents a validation of domestic semiconductor viability for large-scale commercial AI deployments.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.