Skip to main content

Zhipu AI's GLM-5 Leak Sparks Market Frenzy

Zhipu AI's GLM-5 Model Leak Sends Shockwaves Through Tech Sector

The Chinese artificial intelligence industry witnessed an unexpected development during the 2026 Lunar New Year celebrations when details about Zhipu AI's next-generation GLM-5 model surfaced unexpectedly. The leak triggered immediate market reactions, with Zhipu's stock price skyrocketing by an astonishing 200% and pushing its market valuation to HK$150 billion - triple its initial public offering value.

Mystery Model Revealed

Tech enthusiasts first noticed something unusual when a model called "Pony Alpha" appeared on OpenRouter, a global platform for AI services. Its remarkable coding abilities - rivaling those of Claude Opus - quickly made it the talk of online tech communities.

The model's true identity became clear through two telltale signs:

  • System prompts that referenced GLM technology
  • Distinctive quirks characteristic of Zhipu's models, like producing specific unusual responses to certain prompts

Technical Breakthroughs

What makes GLM-5 stand out? The model adopts DeepSeek-V3's efficient sparsity attention architecture (DSA) while making significant improvements:

Feature Improvement

The decision to build on DeepSeek's architecture appears particularly strategic. Not only does it save development resources, but it also allows compatibility with existing optimization frameworks like vLLM and SGLang - potentially saving enterprises millions in deployment costs.

Market Impact

The leak couldn't have come at a better time for Zhipu AI. As Chinese tech firms increasingly compete globally in AI development, demonstrating this level of technical capability sends a strong message to both investors and competitors.

The inclusion of video understanding capabilities specifically addresses what many saw as DeepSeek's primary limitation - its text-only approach. In today's visual-heavy digital landscape, this enhancement could prove crucial for real-world applications.

Key Points:

  • Market Reaction: Zhipu stock surged 200% following the leak
  • Technical Specs: Features DeepSeek-inspired architecture with doubled parameters (745B total)
  • Efficiency: Maintains high performance while activating only small portions of its neural network
  • New Capabilities: Adds robust video understanding to compete in multimedia applications

Enjoyed this article?

Subscribe to our newsletter for the latest AI news, product reviews, and project recommendations delivered to your inbox weekly.

Weekly digestFree foreverUnsubscribe anytime

Related Articles

News

Flapping Airplanes Soars with $180M Funding to Mimic Human Learning

AI research lab Flapping Airplanes has landed a massive $180 million seed round from top investors like Sequoia Capital. Unlike typical AI startups drowning in data, this team wants to teach machines to learn smarter - aiming for human-like efficiency. Founded by brothers Ben and Asher Spector with Aidan Smith, they're betting big on neuroscience-inspired approaches that could revolutionize how AI learns.

February 11, 2026
AI ResearchVenture CapitalMachine Learning
News

iFLYTEK's Xinghuo X2: A Homegrown AI Powerhouse for Specialized Needs

Chinese tech firm iFLYTEK has unveiled its latest AI achievement - the Xinghuo X2 large language model. What sets this apart isn't just its enhanced capabilities, but its complete development on domestic computing infrastructure. Designed for real-world impact, the model specializes in education, healthcare, automotive systems, and intelligent agents. This launch signals China's growing independence in critical AI technologies at a time when computing power has become a strategic resource.

February 11, 2026
AI DevelopmentChina TechSpecialized AI
News

Kimi's Computing Crunch Sparks Unlikely AI Alliance

When Moonshot AI's Kimi chatbot hit computing limits during peak traffic, it did something unexpected - it recommended users try rival DeepSeek instead. This surprising move highlights both the explosive growth of China's AI sector and the industry-wide scramble for computing resources. The candid response won praise online for its transparency amid what analysts call 'growing pains' for domestic large language models.

February 10, 2026
AI IndustryComputing PowerChinese Tech
News

Baidu Quietly Rolls Out 'Project O' Ahead of Lunar New Year

Baidu has discreetly initiated 'Project O,' focusing on enhancing AI capabilities within its flagship app. With three AI platforms already boasting over 100 million users each, this move signals Baidu's commitment to strengthening its position in China's competitive AI landscape. The timing coincides with Baidu's massive Spring Festival campaign, suggesting potential AI-powered features for holiday interactions.

February 10, 2026
BaiduArtificial IntelligenceChinese Tech
Cursor's Composer1.5: A Quantum Leap in AI Coding Assistance
News

Cursor's Composer1.5: A Quantum Leap in AI Coding Assistance

Cursor has unveiled Composer1.5, its most advanced coding assistant yet. The new model boasts a 20x boost in reinforcement learning capacity, delivering smarter responses and tackling complex tasks with unprecedented efficiency. What really sets it apart? A clever 'self-summarization' feature that keeps long coding sessions on track, plus intelligent pacing that knows when to think deep and when to respond fast.

February 10, 2026
AI ProgrammingDeveloper ToolsMachine Learning
Alibaba's Qwen3.5 AI Model Nears Release with Vision-Language Capabilities
News

Alibaba's Qwen3.5 AI Model Nears Release with Vision-Language Capabilities

Alibaba's next-generation AI model Qwen3.5 appears ready for launch, with code appearing in the HuggingFace repository. The model reportedly features a hybrid attention mechanism and may debut as a native vision-language model (VLM). Developers have spotted references to both a compact 2B dense model and a more powerful 35B-A3B MoE variant. If current rumors hold true, Chinese New Year celebrations might coincide with this significant open-source release in the AI community.

February 9, 2026
AIMachine LearningAlibaba