Perplexity has unveiled a research preview of a new orchestrator model for its local AI agent, Perplexity Computer. This system is based on GLM 5.2 from China's Z.ai and has been fine-tuned for operation within the Computer environment.

We're releasing a research preview of a new orchestrator model in Perplexity Computer.

The model is an adapted version of GLM 5.2, post-trained for the Computer harness. It delivers near-frontier performance at 0.344x of the cost of Opus. pic.twitter.com/jcxikoFRfn

— Perplexity (@perplexity_ai) July 9, 2026

"[The model] achieves near-frontier performance at 0.344x the cost of Opus," stated Perplexity.

This is not about completely replacing Claude Opus for all tasks, but rather a hybrid approach. GLM 5.2 handles the majority of requests and escalates to a more powerful model when necessary.

We’ve been post-training a version of GLM that is trained to escalate to a frontier model inside the Computer harness. When paired with an advisor, this model functions at Opus 4.8 grade performance at a fraction of the cost. Available now as a research preview! https://t.co/7y8CjOWOtI

— Aravind Srinivas (@AravSrinivas) July 9, 2026

"In conjunction with an advisor, this model operates at Opus 4.8 level with significantly lower costs," wrote Perplexity CEO Aravind Srinivas.

According to Decrypt, the new system is already available in a research preview format. Full benchmarks from Perplexity are promised in the coming weeks.

The company described the new tool as an adapted version of GLM 5.2, fine-tuned for an agent environment where models perform user tasks through other AI systems. A key component is the advisor tool, which helps determine when capabilities are maxed out and a more powerful model is needed.

Perplexity Computer operates as an agent system that can distribute tasks among multiple AI models. According to Decrypt, the product orchestrates over 19 models, and the adapted GLM is expected to serve as a cost-effective foundational layer for most tasks.

Perplexity's next step is to post-train Nemotron 3 Ultra, another open model the company aims to adapt for Computer.

GLM 5.2 is an open model from Z.ai (formerly Zhipu AI). It contains approximately 744 billion parameters and is released under the MIT license. This allows developers to download, modify, and fine-tune it for commercial products without the restrictions typical of closed APIs.

Z.ai has been on the U.S. Entity List since January 2025.

Decrypt noted the geopolitical context. Perplexity utilizes a model from China's Z.ai but hosts the adapted version in the U.S. on its own infrastructure. This reduces dependence on external APIs and allows the company to control the post-training, routing, and task execution layers.

A similar approach was previously used by Perplexity with DeepSeek R1. In 2025, the company released its adapted version R1-1776, which reportedly removed restrictions on about 300 topics related to Chinese censorship.

In the case of GLM 5.2, Perplexity does not emphasize political neutralization of the model but uses it as an economic layer for the agent system. The goal is to reduce operational costs without completely abandoning advanced models.

In March, Perplexity introduced Personal Computer as a competitor to OpenClaw. It operates 24/7 and retains "memory" between sessions.