Reuters reporting points to a possible August GLM-5.5 launch, but Zhipu AI still hasn't made the model official.
Reuters reported in late June that Z.ai's next GLM-5.5 model is expected in August. A JPMorgan forecast puts the model above one trillion parameters, according to market reports, and the obvious assumption is that Zhipu will keep the 1-million-token context window it shipped with GLM-5.2. Zhipu itself has confirmed none of that. As of August 3, the company's official catalog still stops at GLM-5.2. No GLM-5.5 model card. No benchmark. No date.
The Cadence Is the Only Evidence #
What's actually driving the August guess is arithmetic. GLM-5 arrived in February with roughly 745 billion total parameters, 44 billion active parameters and a 202,000-token context window, according to Z.ai's own GLM-5 materials. GLM-5.1 followed in April. GLM-5.2 landed in June with 753 billion parameters on Hugging Face, MIT-licensed weights and a 1-million-token context window. Run that cadence forward and you land in August.
That's it. It's a pattern, not a promise.
Investors are behaving as if the pattern will hold anyway. Zhipu, now marketing itself as Z.ai and listed in Hong Kong as Knowledge Atlas Technology, went public on January 8 after raising about HK$4.35 billion, around $558 million, in its IPO. Qiming Venture Partners called it the world's first listed large language model company, while Yicai put its debut market value at about $7.4 billion.
The stock then turned into one of the strangest AI trades of the summer. The South China Morning Post reported that Zhipu's market value passed HK$1 trillion, about $128 billion, on June 22 as the shares touched HK$2,980 intraday before closing at HK$2,410. SCMP said the stock had risen more than 1,700% since its January debut. The same JPMorgan note raised Zhipu's 2026 to 2030 revenue forecast by 7% to 16%, lifted its price target from HK$950 to HK$1,400, and projected 2026 revenue growth above 534%, SCMP reported.
Bigger Isn't the Bet #
Here's the thing that undercuts the hype. A trillion-plus parameters sounds massive until you look at what China's other labs shipped in July. Moonshot AI announced Kimi K3 on July 16 with 2.8 trillion parameters, and Tom's Hardware reported that the weights were released six days before August 3. Alibaba previewed Qwen3.8-Max on July 19, a 2.4-trillion-parameter multimodal model. No benchmark table, no weights at launch. DeepSeek's own V4 documentation lists DeepSeek-V4-Pro at 1.6 trillion total parameters with 49 billion active.
Against that field, a one-trillion-parameter GLM-5.5 wouldn't win on size. Don't pretend otherwise.
Zhipu's stronger case is software engineering. GLM-5.2's official Hugging Face card reports 62.1% on SWE-bench Pro against GPT-5.5's 58.6%, a useful edge on a benchmark built around fixing real software issues rather than answering trivia. Claude Opus 4.8 still leads GLM-5.2 on Terminal Bench 2.1, 85.0% to 81.0%, and on SWE-bench Pro at 69.2%. Zhipu isn't claiming to beat Anthropic outright. Its pitch is that it gets close enough, with open weights and lower serving costs, for developers who can't treat Claude pricing as a rounding error.
That cost gap is the real story underneath the parameter count. If Zhipu ships an open-weight GLM-5.5 with better coding scores and keeps the economics close to GLM-5.2, you have a serious new option for coding agents. If you're budgeting inference costs against Claude or GPT-5.5 this quarter, that matters more than another round number in a launch deck.
An unconfirmed model is still an unconfirmed model. Zhipu has launched quickly before, and its silence doesn't prove anything either way. The next real evidence will be dull and useful: a model card, a repository, an API entry or a pricing page. As of August 3, there still isn't one.
Also read: EU AI Act Transparency Rules Take Effect Today Despite the 2027 Delay • Anthropic's COBOL Tool Rattled IBM, But AI Migration Still Hides Silent Bugs • Retail Traders Are Building AI Trading Bots Once Reserved for Hedge Funds