According to Perplexity's announcement on July 9, the company released a post-trained version of GLM 5.2, a 744-billion-parameter Chinese open-source model, designed to operate inside its Computer agent harness. The fine-tuned model delivers near-frontier performance at approximately one-third the cost of Claude Opus 4.8 (0.344x).
The system uses an "advisor tool" that recognizes when a query exceeds its own capabilities and escalates to a frontier model for complex tasks, keeping costs low by handling most tasks independently. CEO Aravind Srinivas noted that when paired with the advisor, the model functions at Opus 4.8-grade performance for a fraction of the price. The model is available now as a research preview, with full benchmarks and research paper expected in coming weeks.