Perplexity Introduces GLM 5.2 Model
Perplexity has rolled out a research preview of its newly fine-tuned AI model, GLM 5.2, designed to integrate with its Computer system and serve as a more economical alternative to the Claude Opus 4.8 model. This adaptation allows the company to operate at one-third the cost of Opus 4.8 across various benchmarks.
Details on GLM 5.2 Adaptation
The refined version of GLM 5.2, originating from Z.AI (formerly Zhipu AI), is characterized by approximately 744 billion parameters. Released under an MIT license, this model is noted for its strong capabilities in coding tasks while significantly reducing API costs. Perplexity’s previous fine-tuning of a Chinese model was R1-1776, which saw modifications to meet compliance by omitting certain topics censored in China.
Fine-Tuning and Performance Insights
Fine-tuning leverages a trained AI model and optimizes it on specific datasets to enhance its performance. In this instance, Perplexity applied post-training methods to equip GLM 5.2 with a unique "advisor tool" that determines whether a task should be handled by itself or escalated to a more powerful model. CEO Aravind Srinivas noted, "This model performs at Opus 4.8 grade quality for a fraction of the cost." This integration streamlines costs, achieving operational expenses around twice that of the base model but significantly lower than relying exclusively on the Opus 4.8 model, which is over 600% more expensive for the same tasks.




