Perplexity has released a research version of a new orchestrator model for its local AI agent, Perplexity Computer. The system is based on GLM 5.2, an open-source model from the Chinese company Z.ai (formerly Zhipu AI), which has been fine-tuned specifically for operation in the Computer agent environment.
Cost Efficiency Without Quality Loss
The key advantage of the new architecture is a radical reduction in cost. As noted by Perplexity, the adapted GLM 5.2 delivers performance "on the edge of cutting-edge achievements" at a cost of just 0.344x that of using Opus. This is not about completely replacing top-tier models, but about a hybrid scheme: GLM 5.2 handles the bulk of requests, and when necessary, passes complex tasks to more powerful models via a built-in advisor tool.
Perplexity CEO Aravind Srinivas emphasized that, in conjunction with the advisor, the system operates at the level of Opus 4.8, but at significantly lower costs. This makes the solution particularly attractive for commercial use, where every query matters.
Technical Details and Geopolitical Context
GLM 5.2 contains approximately 744 billion parameters and is released under the MIT license, allowing developers to freely download, modify, and fine-tune the model without the restrictions typical of closed APIs. Perplexity hosts the adapted version on its own infrastructure in the United States, reducing dependence on external APIs and providing full control over the post-training, routing, and task execution layers.
This approach is particularly interesting against the backdrop of geopolitical tensions. Z.ai has been on the U.S. Entity List since January 2025, but Perplexity uses the model not as a political tool, but as an economic layer for its agent system. The goal is to reduce operational costs without completely abandoning advanced models. The company previously employed a similar strategy with DeepSeek R1, releasing an adapted version called R1-1776.
What's Next?
Perplexity Computer operates as an agent system orchestrating over 19 models. The adapted GLM 5.2 is intended to serve as a low-cost base layer for most tasks. The company's next step is the post-training of Nemotron 3 Ultra, another open-source model for Computer. Full benchmarks for the new system are promised for release in the coming weeks.
Expert Commentary: This move by Perplexity is a vivid example of how open-source models are beginning to displace expensive proprietary solutions in the AI agent niche. The hybrid architecture with task escalation is not just about cost savings, but a new paradigm where efficiency is achieved through intelligent load distribution, rather than a race for raw computing power. If the benchmarks confirm the stated figures, we can expect a wave of similar solutions from other players.