A Defining Moment for Open-Source AI

The landscape of artificial intelligence has a new frontrunner. Zhipu AI's latest flagship model, GLM-5.2, has achieved a top score of 51 points in a widely-respected benchmark evaluation, decisively claiming the number one position among all open-source large language models globally.

Performance That Rivals the Best

The model's capabilities shine even brighter in tests simulating complex, real-world knowledge work. Scoring an impressive 1524 points against a human benchmark of 1000, GLM-5.2's performance not only surpasses other open-source alternatives but places it on par with the most advanced proprietary models available today, signaling a major shift in the competitive balance.

Significant Gains Across the Board

Compared to its predecessor, GLM-5.2 demonstrates substantial improvements in key areas:

  • Scientific Reasoning: A 16-percentage-point increase in critical evaluation scores.
  • Higher-Level Reasoning: Performance boosted by 12 percentage points.
  • Comprehensive Benchmarking: Achieved a 16-percentage-point gain.
  • Expert-Level QA: Secured a high score of 89% on an exceptionally challenging test.

Optimal Balance of Intelligence and Cost

GLM-5.2 occupies the most cost-effective position on the performance-versus-cost curve. While the average cost per task has increased due to its significantly longer output length (averaging 43,000 units), it remains far more economical than proprietary models of similar capability, offering a powerful open-source alternative for developers and enterprises.

Technical Specifications and Availability

Boasting 744 billion total parameters with 40 billion activated per inference, the model features a massively expanded context window of 1 million tokens—a substantial upgrade from 200k—greatly enhancing its ability to process lengthy documents and complex dialogues. Released under the permissive MIT license, the model is fully open-source.

Access to GLM-5.2 is now available through Zhipu AI's official API service, featuring competitive pricing, as well as through several leading cloud and AI service platforms, enabling rapid integration and deployment for a wide range of applications.