Zhipu's Tang Jie Internal Letter 'The Giant Wave Has Come': Launches Two-Year 'Touch High' Plan to Aim for AGI and ASI

On July 11, 2026, Zhipu founder Tang Jie released an internal letter titled 'The Giant Wave Has Come', systematically elaborating the company's underlying judgment on AI transformation and its new strategy post-IPO. The letter announced the launch of a two-year 'Touch High' plan, focusing on four engines: long-horizon tasks, autonomous agent systems, self-evolution, and extreme safety governance. It temporarily suspends short-term commercialization to fully pursue Artificial General Intelligence (AGI) and even Artificial Superintelligence (ASI).

Background: IPO Reset, Counter-Intuitive Strategy

Zhipu listed on the Hong Kong Stock Exchange on January 8, 2026, with an issue price of HKD 116.2. In the following six months, the stock price peaked at HKD 2980, a more than 24-fold increase from the issue price, with a market cap once exceeding HKD 1.3 trillion, surpassing Baidu and Xiaomi. After the first batch of shares unlocked on July 8, the stock price rose instead of falling. The next day, Zhipu placed new shares at HKD 1588 per share, raising approximately HKD 31.4 billion for foundational model R&D and computing infrastructure. Tang emphasized in the letter that on the day of listing, the company decided to 'reset' and fully return to foundational model research, believing that 'true commercial opportunities lie in the leap of the intelligence upper bound.'

Core Strategy: Touch High Plan

Tang proposed that reaching AGI requires scaling three technical peaks:

• Long-Horizon Tasks: Enable models to plan and execute tasks spanning weeks to months, developing a new generation of memory architecture.
• Autonomous Agent Systems: Build thousands of 'digital employees' operating 7×24 autonomously and collaboratively, achieving a no-person company (NPC). The three major challenges of memory, continuous learning, and self-judgment have been progressively resolved in engineering.
• Self-Evolution: AI training AI has taken shape; models can write code, synthesize data, and train themselves, forming a recursive improvement flywheel.

Additionally, Zhipu plans to invest billions of yuan to tackle 'extreme safety governance', advancing model interpretability research and embedding ethical norms at the foundational level.

Technical Progress and Commercial Achievements

Zhipu's flagship model GLM-5.2 (released in June 2026) matches or surpasses Claude Opus 4.8 and GPT-5.5 on multiple core benchmarks, is open-sourced under the MIT license, and supports 1M context. The 2025 financial report shows that the MaaS platform ARR reached RMB 1.7 billion, a 60-fold increase over the past year. Tang noted that betting on coding and reasoning capabilities was key to Zhipu's takeoff.

Prediction on ASI

Tang cited a Google DeepMind report: even if individual model capabilities stagnate, if AGI instances expand tenfold annually, one hundred million agents sharing a common underlying brain would collectively equate to ASI in five years. He believes AI will gradually touch self-awareness, emotion, and even consciousness—a trend that is irreversible and will reshape application layers, operating systems, and even the von Neumann architecture.

Industry Reactions and Significance

The internal letter sparked widespread discussion in the industry. Zhipu's counter-intuitive return to foundational research at the peak of commercialization is interpreted as a long-term bet on the AGI endgame. Tang wrote at the end of the letter: 'Not reaching the summit means failure.'