Zhipu CEO Tang Jie’s Internal Memo: Beyond the GLM Moment
Zhipu founder and CEO Tang Jie releases landmark internal memo titled “The Wave Has Arrived” on July 11, 2026.
🌊 Market Milestone: From Startup to Trillion-HKD Powerhouse
On July 11, 2026, Zhipu AI’s founder and CEO Tang Jie issued a pivotal internal memo titled “The Wave Has Arrived”, marking a defining inflection point in the company’s trajectory.
In just six months:
– Zhipu’s market capitalization surged 10× from its H-share listing debut in January 2026;
– By June 2026, it joined the elite “Trillion-HKD Club”, with a valuation exceeding HK$1 trillion — nearly 3× Baidu’s market cap, and surpassing Xiaomi;
– Despite the first tranche of share lockup expirations on July 8, Zhipu maintained strong price stability — signaling deep investor confidence in its strategic direction.
This meteoric rise reflects the most compelling narrative in today’s large language model (LLM) landscape: betting correctly on foundational technical direction yields unmatched commercial traction and global recognition.
⚙️ The Coding Bet: How Zhipu Rewrote the Rules
Zhipu’s ascent traces back to a bold, counterintuitive pivot in early 2025: a full-scale strategic reorientation toward AI coding capability.
At a public event, Professor Tang Jie — also Head of Computer Science at Tsinghua University — articulated the rationale:
“The emergence of DeepSeek R1 signals the end of the Chat paradigm. In the post-DeepSeek era, we ‘bet’ on two pillars: Coding and Reasoning — capabilities designed for symbiotic coexistence with autonomous agents.”
That bet paid off decisively:
– GLM-4.5 (released July 2025) laid the groundwork;
– GLM-5.2, open-sourced in June 2026, now ranks among the global top-tier AI coding models, matching or outperforming Claude Opus 4.8 and GPT-5.5 across key benchmarks;
– Revenue followed: Zhipu’s MaaS platform achieved ¥1.7B ARR (as of March 2026), a staggering 60× increase year-on-year.
🧭 Three Strategic Pillars for the Post-Coding Era
Tang Jie’s memo declares that the “GLM moment” is over — and introduces Zhipu’s next-generation mission: scaling the intelligence frontier. He identifies three interlocking technical mountains to climb:
🔹 Long Horizon Task (LHT)
Enabling models to plan, execute, and adapt across weeks, months, or years — not just instant answers. Example: an AI continuously auditing software for zero-day vulnerabilities, emulating and amplifying expert human cognition.
🔹 Autonomous Agent System
Building self-organizing, collaborative swarms of intelligent agents — moving beyond “assistants” to true digital employees. Core enablers already advancing:
– Memory: via extended context + RAG;
– Continual Learning: accelerated model iteration cycles;
– Self-Judgment: emerging in frontier LLMs.
🔹 Self-Evolving AI
Where models train themselves: writing code, synthesizing training data, and iteratively refining architectures — all within secure sandboxes. As Tang notes: “Speed is the ultimate moat; rapid iteration creates generational cognitive gaps.”
🚀 The ‘Touch High’ Strategic Plan
To operationalize this vision, Zhipu launches “Touch High” — a two-year, AGI-focused initiative prioritizing fundamental breakthroughs over short-term monetization.
Four core engines drive the plan:
| Engine | Focus Area | Key Initiative |
|---|---|---|
| 1 | Long Horizon Task | Next-gen memory architecture enabling full-lifecycle learning & autonomous decomposition of grand challenges (e.g., designing novel anticancer molecules) |
| 2 | Autonomous Agent System | A society of thousands of specialized agents — debating, reviewing code, allocating resources — achieving autonomous digital productivity |
| 3 | Fully Self-Training | AI-powered synthetic data factory + self-play adversarial training + safe in-sandbox code evolution |
| 4 | Extreme Safety & Governance | Embedding ethics, laws, and social norms into the model’s value function; billion-RMB investment in mechanistic interpretability; active participation in global AI governance |
💡 “Safety is not a feature — it’s the precondition for existence. When intelligence reaches civilizational scale, alignment must march in lockstep with capability.”
🌐 Openness as a Strategic Imperative
True to its philosophy, Zhipu released GLM-5.2 under the permissive MIT License, supporting 1M-context windows, and made it fully downloadable, deployable, and commercially usable — no restrictions, no gatekeeping.
As Tang affirms:
“We reach up to touch the highest peaks — while simultaneously building roads down for everyone. The height belongs to humanity. The road belongs to every developer.”
✨ Closing Vision: Why Now? Why Zhipu?
“Why push upward when others accelerate commercialization? Because true pioneers don’t just climb mountains — they turn them into highways.
We are building that highway — high enough to safeguard national security and human dignity; wide enough for every team, every student, every startup to ascend.
This is our generation’s greatest privilege — and gravest responsibility.
The wave has arrived. Zhipu will ride it — upward, unflinchingly.
— Tang Jie, Founder of Zhipu AI
July 11, 2026
Source: Originally published by “Intelligence Emergence” WeChat Official Account.