Zhipu Surges Over 8% as Next-Gen Model GLM-5.3 Allegedly Leaked Ahead of Schedule
0xBroomberg
Zhipu (02513) surged over 8% in Hong Kong trading to HK$1,015, after technical documentation for its next-generation model GLM-5.3 was briefly and apparently accidentally made public — the market is front-running the bet that the next model leapfrogs the last.
Why did the stock spike?
Zhipu shares jumped over 8% intraday on August 4, hitting HK$1,015 with turnover of HK$1.27 billion.
The trigger: on the afternoon of August 3, users discovered that GLM-5.3's official documentation page was briefly accessible — it was open for roughly one hour before being taken down.
This means → the market read a suspected document leak as a signal that a new model launch is imminent, and money moved in ahead of any official announcement.
What is GLM-5.3, and why does it matter?
GLM-5.3 is the next-generation flagship model Zhipu is developing. Its predecessor, GLM-5.2, launched just this June.
GLM-5.2 scored 51 points on the Artificial Analysis composite benchmark — a third-party AI model evaluation platform — trailing only Anthropic and OpenAI, the highest score among open-source models.
In plain terms = if 5.2 is already the top dog in the open-source camp, whether 5.3 can hold or widen that lead is the single question the market cares about most.
What do brokers say about Zhipu's outlook?
China Securities International previously said it is bullish on the GLM model family leading Chinese AI models to gain competitiveness and market share in global large-model competition.
The brokerage sees Zhipu's commercialization as having potential for "exponential growth."
This means → the sell side has already positioned Zhipu as a frontrunner for China's AI model expansion globally; GLM-5.3's performance will be the direct test of whether that thesis holds.
What does this "leak" tell us — and what doesn't it?
What it tells us: Zhipu is actively building the next-generation model. The documentation page was already set up, suggesting a formal release may not be far off.
What it doesn't: the page was only briefly accessible and disclosed no specific performance benchmarks — how much better 5.3 is than 5.2 remains unanswered.
This reflects the market's acute sensitivity to AI model iteration speed: even an accidental one-hour document exposure was enough to drive a single-day rally of over 8%.
Content is for reference only, not financial advice.