Programming AI changed again? Mysterious model Pony Alpha goes viral, rumored to be Zhipu GLM-5, stock price skyrockets 60% in two days!

```

Driven by news that the mysterious high-performance model “Pony Alpha” is suspected to be the new flagship product from Zhipu AI, the company's Hong Kong stock price has surged sharply for two consecutive trading days. The market speculates that the model codenamed “Pony Alpha” is actually Zhipu’s soon-to-be-released GLM-5. With strong market expectations, Zhipu AI’s stock price has risen 60% in two days, setting a historic high.

According to Yicai, citing insiders, Zhipu AI currently has a relevant confidential project underway, and the mysterious model sparking heated discussions on the OpenRouter platform is highly likely to be the upcoming GLM-5. Boosted by this sentiment, Zhipu AI’s market value briefly surpassed HKD 150 billion, nearly three times its IPO value, showing investors’ high expectations for domestic large model technology iterations.

The catalyst for this round of market activity originated from the anonymous model codenamed “Pony Alpha” launched on the global model service platform OpenRouter on February 6. Although the model was released in “stealth mode” with no developer indicated, its coding, reasoning, and agent workflow abilities have been described by the developer community as having “Opus-level” performance. According to OpenRouter partner Kilo Code, the model is “a special evolution of the world’s most popular open-source model in a global lab,” further fueling market speculation about its background.

As more technical details emerged, multiple chains of evidence started pointing to Zhipu AI. From the system prompt test in which the model identifies as “I’m GLM”, to tokenization performance identical to GLM-4, and hints from Zhipu’s chief scientist about “the release of a new model around the Spring Festival”, market consensus has gradually formed. This unusual stock movement reflects the capital market’s renewed assessment of the global competitiveness of domestic large models in advanced coding and complex logical reasoning.

Core Evidence Points to GLM-5

Regarding the true identity of “Pony Alpha”, the tech community has unearthed key clues through multidimensional reverse testing.

According to media reports, while the OpenRouter page only shows that the model supports a 200K context window and up to 131K output, in system prompt modification tests the model directly responds with “I’m GLM.” In addition, developers using specific strings for PoC token testing found that the model’s response at the tokenizer level was identical to GLM-4, and its code formatting style was highly consistent with Zhipu models.

Besides software features, open-source communities discovered architectural details from GitHub code commits that further corroborate GLM-5’s technical path. Developers found in vLLM inference framework PRs that GLM-5 adopts DeepSeek-V3/V3.2 architecture, including sparse attention mechanism (DSA) and multi-token prediction (MTP) technology. According to code inferences, GLM-5 has 745B parameters, 78 hidden layers, uses a MoE (Mixture of Experts) architecture with 256 experts, and activates 8 with each inference. This architectural change means GLM-5 maintains high performance while benefiting directly from optimization by inference frameworks like vLLM, greatly lowering deployment barriers.

Programming and Agent Capabilities Go Viral

In actual application tests, the engineering capabilities demonstrated by “Pony Alpha” are the core reason for the market’s attention. Media reviews found the model performed strongly in coding, reasoning, and role play, and is considered a next-generation flagship foundational model on par with Claude Opus. In front-end development tests, the model can generate a complete broadcasting application with over 500 lines of code from a single prompt, with UI design and interactive logic reaching mature product standards.

Even more outstanding is its Agentic Coding capacity. In tests recreating complex game projects like “Stardew Valley”, the model displayed “architect thinking”, autonomously breaking down system requirements, planning project architecture, and continuously programming for over 10 minutes from frontend rendering to backend database construction. Additionally, in legacy financial system code refactoring tests, the model successfully modernized the code while retaining key business logic, showing very high usability and stability.

Spring Festival AI “Arms Race” Upgrades

The potential release time for Zhipu GLM-5 coincides with the intensive “Spring Festival” batch release period for domestic large models. Zhipu AI’s chief scientist Tang Jie previously previewed in an internal letter that GLM-5 would be released, pointing to mid-February 2026. Meanwhile, new models from DeepSeek, the MiniMax planned M2.2, and Qwen 3.5 are also reported to be updated soon.

OpenRouter data shows that over 91% of community users believe “Pony Alpha” is a test version of GLM-5. If the model is ultimately confirmed as GLM-5, it will mean Zhipu has achieved a technical breakthrough on the basis of reusing DeepSeek’s efficient architecture, and marks a new stage of competition for domestic foundational models in advanced programming and engineering agents.

For investors, this technological leap directly translates into a revaluation of Zhipu AI, and the market is closely watching the company’s official release and subsequent commercialization.

Risk Warnings and DisclaimerThe market has risks, investment needs caution. This article does not constitute individual investment advice, nor does it take into account individual users’ special investment goals, financial situation, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are appropriate for their specific situations. Invest accordingly at your own risk. ```