Perplexity Drops a $0.34-on-the-Dollar Claude Imposter — and It's Cheaper Than Your Cloud Bill 💸

Perplexity on July 9, 2026 released a research preview of a post-trained version of Z.AI's GLM 5.2, an open-source model it said operates at near-frontier performance for roughly 0.344x the cost of Anthropic's Claude Opus 4.8. The adapted model is built to run inside Perplexity's Computer agent harness and is available now in production, the company said.

GLM 5.2 is a roughly 744-billion-parameter model from Z.ai, formerly Zhipu AI, a Beijing-based lab that has been on the U.S. Entity List since January 2025. Released under an MIT license in June, the model posts top-tier scores on long-horizon coding benchmarks and can be downloaded, modified and fine-tuned for commercial use without restriction. Perplexity applied post-training to teach the model when to handle a task itself and when to hand it off to a more capable system.

The fine-tuned GLM 5.2 includes what Perplexity calls an "advisor tool," a native capability that recognizes when a query exceeds its own competence and routes the request to a third-party frontier model. "When paired with an advisor, this model functions at Opus 4.8 grade performance at a fraction of the cost," CEO Aravind Srinivas wrote on X. The architecture is designed to keep the majority of inference traffic on the cheaper model, with the expensive frontier system engaged only on tasks that require it.

Perplexity benchmarked the system against the standard GLM 5.2 to establish a cost baseline, using an internal efficiency metric that measures the dollars required to complete complex tasks. The combined cost figure of approximately one-third of Opus 4.5 was reported in the company's announcement materials. The company did not disclose which frontier model serves as the escalation target inside the Computer harness, and the research preview is positioned as a step toward broader deployment of the orchestrator pattern across Perplexity's product surface.