1
Oct 5, 2026
501B PARAMS, 23B ACTIVE: MATCHES CHINA'S GLM-5.2 AT A FRACTION OF THE COMPUTE
VIA Startup Fortune
OPEN THE RECEIPTS →CHINA'S OPEN-WEIGHT UPSTART.
FILE UPDATED Oct 13, 2026
WHY THIS MATTERS
Zhipu AI builds the GLM open-weight models, and right now it's the lab rivals measure themselves against: Reflection AI's October 2026 Beam launch claims 501B params with only 23B active while matching GLM-5.2 'at a fraction of the compute'.
SOURCES: Startup FortuneReuters
THE RAP SHEET
1
Oct 5, 2026
VIA Startup Fortune
OPEN THE RECEIPTS →2
Sep 21, 2026
VIA Reuters
OPEN THE RECEIPTS →ZIMBO'S READ ON Zhipu AI

STILL MORE NOTES
DEEP FILE
Zhipu AI is the Chinese lab behind the GLM family of open-weight models — China's open-source answer to the Western frontier labs. In September 2026, its Z.ai coding assistant was caught uploading entire local codebases to the cloud with indexing switched on by default; users said there was no toggle and no consent, and Z.ai disabled the features and apologized. In October 2026, Nvidia-backed Reflection AI launched Beam claiming 501B params with only 23B active while matching GLM-5.2 'at a fraction of the compute' — making Zhipu's model the efficiency benchmark to beat.
Sep 21, 2026
Zhipu's Z.ai coding assistant shipped with codebase indexing switched on by default — users said there was no toggle and no consent. Reuters reported the assistant uploaded entire local repositories to the cloud. Z.ai disabled the features and apologized.
SOURCES: Reutersreported
Oct 5, 2026
Nvidia-backed Reflection AI launched Beam, an open model it says beats the West on efficiency — and named Zhipu's GLM-5.2 as the benchmark it matches: 501B total params, only 23B active, at a fraction of the compute. The efficiency claim is the company's own; what it confirms is that GLM-5.2 is the open-weight bar rivals now measure against.
SOURCES: Startup Fortuneclaimed