Back to timeline

Sat, July 1119:28Model/APIModel releasesAI codingChinese modelsModel releases guide

Zhipu Founder Tang Jie's Internal Letter: Betting on Long-Horizon Tasks, Autonomous Agents, and Self-Evolution in the Post-Coding Era

Decision Brief

What changedIn an internal letter, Zhipu founder Tang Jie announced a strategic shift to focus on long-horizon tasks, fully autonomous agent systems, and self-evolution, launching the 'Touch High' initiative.
Why it mattersZhipu's pivot from coding to next-gen technologies signals key capability changes in its open-source GLM-5.2 and future models, impacting developers building complex task workflows.
Who should careTeams building on model APIs
Affected stackClaudeOpenAIDeepSeekGLM
Builder actionWorth evaluating: can Claude replace or complement your current model
Source confidenceMedium · Reliable media or first-hand reporting

On July 11, 2026, Tang Jie released an internal letter 'The Great Wave Is Coming', reviewing Zhipu's coding bet: market cap 10x, joining the trillion-HKD club, and MaaS ARR reaching 1.7 billion RMB (60x in a year). The flagship GLM-5.2 is now open-source, supporting million-token context, with performance matching Claude Opus 4.8 and GPT-5.5. Tang noted the chat paradigm is over, and after the DeepSeek era, Zhipu bet on coding and reasoning, now entering a new phase. The company launches the 'Touch High' plan, strategically investing in four core engines over two years: long-horizon tasks (novel memory architectures), autonomous agent systems (building agent societies), fully self-training (synthetic data and self-play), and extreme safety governance. Zhipu will continue its open ecosystem; the latest GLM-5.2 uses the MIT license for anyone to download, deploy, and commercialize. For developers using Zhipu models or open-source GLM-5.2, future models will prioritize long-horizon tasks and autonomous agent capabilities, but commercialization pace may slow short-term. For teams focused on AI coding and agent infrastructure, Zhipu's shift from 'instant Q&A' to 'grand engineering' makes model memory and autonomous task decomposition key competitive edges.

Summary basis: official / RSS sourceCompiled from the source scope noted above; the original remains authoritative.

Sources

  • 36氪

    Chinese tech/startup coverage; AI signal filtered from broader business news.

  • 36氪

Related intel

留言

登入后即可留言,和其他 builder 交换实测心得。

还没有留言,抢头香。