BEIJING, February 12, 2026, 17:07 (GMT+8)
- Zhipu bumped up prices on its GLM coding subscription by at least 30%, with the hike taking effect right away.
- Zhipu just rolled out its open-source GLM-5 model, joining the wave of Chinese AI companies hurrying to debut fresh releases before the Spring Festival holiday.
- Premier Li Qiang urged a broader, commercial rollout of AI, pushing for more unified management of power and computing resources.
Zhipu, the Chinese AI startup, jacked up prices for its GLM coding plan subscriptions by a minimum of 30% after seeing a surge in users leaning on its AI models for coding help. The new rates kick in right away, though current subscribers are spared, according to a post from the company on its WeChat account.
Timing is key here. China’s senior officials are urging companies to move AI beyond prototypes and into real-world, high-volume use—which comes down to electricity, data centres, and customers actually paying. On Wednesday, Premier Li Qiang called for sharper coordination on power and computing, telling state broadcaster CCTV the country needs to accelerate “scaled and commercialised application” of AI. reuters.com
It’s been a year since DeepSeek rattled the sector with its bargain-priced model, and now competitors are gearing up to roll out their own releases just as the Spring Festival kicks off on Feb. 15. “The surprise would be if some of these new models end up being underwhelming. I think there are high expectations here,” said Alfredo Montufar-Helu, managing director at Ankura Consulting in Beijing. Chinese AI model operating costs are coming in at about one-sixth to one-fourth of similar U.S. systems, according to a RAND report from last month. Omdia’s Lian Jye Su points to DeepSeek as proof that solid models can still be built with limited resources. reuters.com
Zhipu rolled out GLM-5 this Wednesday, touting the open-source model’s improved coding chops and its ability to run lengthy “agent” assignments—multi-step tasks wrapped up with minimal prompting. The company claims its benchmarks nearly match Anthropic’s Claude Opus 4.5, and on some tests, it edges out Google’s Gemini 3 Pro. Notably, Zhipu relied on domestic chips for inference—Huawei’s Ascend, as well as hardware from Moore Threads, Cambricon, and Kunlunxin—as U.S. restrictions tighten around advanced semiconductors. MiniMax, a rival, said it launched its own M2.5 open-source model through its overseas agent portal. Both firms listed in Hong Kong last month. Zhipu, for its part, says it’s eyeing global markets despite ongoing U.S. sanctions. reuters.com
Zhipu, trading under the name Knowledge Atlas Technology in Hong Kong, popped 30% Thursday following the GLM-5 debut, according to Seeking Alpha.
Zhipu’s z.ai brand, according to VentureBeat, is positioning GLM-5 as a model with notably low “hallucination” rates—industry jargon for when AI invents convincing but incorrect responses. The company is also rolling out an “Agent Mode,” which lets users spin up files like Word docs, PDFs, and spreadsheets. GLM-5 runs on a “mixture-of-experts” architecture and is distributed under the MIT open-source license. Pricing gets posted on OpenRouter, listed per token—the basic unit for text processing in these AI models. venturebeat.com
Still, there’s room for the narrative to shift. Certain performance boasts depend on company-curated benchmarks, and questions remain over how fast coding assistants and agent tools can actually drive lasting subscription income—particularly with competitors frequently slashing prices. On top of that, U.S. chip restrictions and Zhipu’s potential for sanctions exposure complicate things further, especially as these firms aim for sales outside China.
Now, investors want to see if those model launches actually lead to deals—and whether the government’s AI push is turning into contracts instead of just buzz. The holiday release window? It’s a stress test of its own: lots of hype, little time, and even less tolerance for slip-ups.
