🇨🇳Zhipu Releases GLM-4.7 Trained Entirely On Huawei Ascend Chips, Claims Lowest Hallucination Rate
TL;DR
Chinese AI lab Zhipu released GLM-4.7, a frontier-class model trained end-to-end on Huawei Ascend silicon — not Nvidia. Zhipu claims a 1.2% hallucination rate, the lowest reported by any frontier lab, at $0.11 per million input tokens versus Claude Opus at $15.
Chinese AI lab Zhipu released GLM-4.7, a frontier-class model trained end-to-end on Huawei Ascend silicon — not Nvidia. Zhipu claims a 1.2% hallucination rate, the lowest reported by any frontier lab, at $0.11 per million input tokens versus Claude Opus at $15.
Key Points
Trained entirely on Huawei Ascend silicon, sidestepping US export controls
Claimed 1.2% hallucination rate, lowest reported among frontier labs
Priced at $0.11 per million input tokens
Why It Matters
GLM-4.7 is proof that China's domestic chip stack can train frontier models without Nvidia — a turning point for export controls and global AI supply chains.
Quick Facts
Frequently Asked Questions
Why does this matter?
GLM-4.7 is proof that China's domestic chip stack can train frontier models without Nvidia — a turning point for export controls and global AI supply chains.
What happened?
Chinese AI lab Zhipu released GLM-4.7, a frontier-class model trained end-to-end on Huawei Ascend silicon — not Nvidia. Zhipu claims a 1.2% hallucination rate, the lowest reported by any frontier lab, at $0.11 per million input tokens versus Claude Opus at $15.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,462 builders reading daily.