Perplexity has released a research version of a new orchestrator model for its agent platform, Perplexity Computer. It is based on the open-source Chinese model GLM 5.2 from Z.ai, which has undergone additional post-training for operation in an agent environment. This is not just another AI release, but a well-thought-out step to reduce operational costs without sacrificing quality.
The key innovation is a hybrid architecture. GLM 5.2 handles the bulk of routine queries, and when it reaches the limits of its capabilities, it automatically delegates the task to a more powerful advisor model. According to Perplexity, this combination delivers performance on par with Opus 4.8 at a cost of only 0.344x that of the reference model. In an industry where every cent counts, this is a serious competitive advantage.
The system is already available as a research preview. The company promises to publish full benchmarks in the coming weeks. It is important to understand: this is not about completely replacing top-tier models. GLM 5.2 is a cost-effective base layer that allows for load distribution among 19 models within Perplexity Computer.
The geopolitical aspect is also noteworthy. Perplexity uses the Chinese open-source model Z.ai, which has been on the U.S. Entity List since January 2025. However, the company hosts an adapted version on its own infrastructure in the United States, fully controlling the post-training and routing layer. This reduces dependence on external APIs and provides full control over task execution. Perplexity has previously applied a similar approach with DeepSeek R1, releasing its adapted version R1-1776, stripped of censorship restrictions.
GLM 5.2 contains approximately 744 billion parameters and is distributed under the MIT license, allowing developers to freely modify it. Perplexity's next step is post-training Nemotron 3 Ultra, another open-source model for Computer.
Expert opinion: Perplexity demonstrates a mature approach to building agent systems. Instead of relying solely on expensive proprietary models, the company skillfully uses open-source solutions as an economic foundation, retaining control over critical layers. This is not just cost savings—it is strategic independence from API providers and a foundation for scaling agent computing in the real-world sector.