Zhipu AI Launches Open-Source GLM 5.2 Model with 744 Billion Parameters, Rivals US Frontier Models

Recently, Zhipu AI officially released the open-source GLM 5.2 large model with 744 billion parameters. The model surpasses GPT-5.5 on multiple programming benchmarks, with performance approaching Claude Opus 4.8, while being fully open-sourced under the MIT license and offering API pricing 85% lower than mainstream competitors. This release quickly garnered industry attention, with international tech programs like All In Podcast dedicating special discussions to it.

Technical Breakthrough: Distillation Strategy Enables Rapid Catch-Up

The core competitiveness of GLM 5.2 lies in its unique "Device Farm Distillation" technology. By leveraging a large-scale parallel inference cluster, Zhipu AI uses outputs from US frontier models as training signals to achieve knowledge transfer in a short time. This approach significantly reduces the cost of training from scratch while maintaining high performance levels. Test data shows the model outperforms GPT-5.5 by approximately 3–5 percentage points on coding tasks such as HumanEval and LiveCodeBench.

Open-Source Strategy and Price Advantage

Unlike most closed-source models, GLM 5.2 is released under the permissive MIT license, allowing commercial use and derivative works. The API pricing strategy is equally aggressive, offering 85% lower costs compared to models of similar scale, providing a more accessible entry point for small and medium-sized enterprises and research institutions. The open-source community responded quickly, with the relevant GitHub repository garnering over 20,000 stars within 24 hours.

Industry Impact and Ecosystem Outlook

Analysts point out that the release of GLM 5.2 signals a shift for open-source models from "followers" to "parallel runners." While the distillation method has sparked discussions around intellectual property, it has objectively accelerated the diffusion of global AI technology. In the future, the open-source ecosystem may develop stronger competitiveness in areas such as coding assistance and scientific research simulation.

Conclusion

The debut of GLM 5.2 not only showcases the technical strength of the Chinese AI team but also injects new variables into the global open-source large model competition. As more players join, the boundary between open-source and closed-source camps is likely to become increasingly blurred.