Contact:
info@dailynewstech.com
seo@dailynewstech.com
pr@dailynewstech.com
ads@dailynewstech.com
Perplexity Launches New Orchestrator Model for Computer: GLM 5.2-Based, at a Third of Opus’s Cost
Perplexity has released a research preview of a new orchestrator model inside Perplexity Computer, the company announced in a LinkedIn post. The headline claim: near-frontier performance at a fraction of the price of a frontier model.
What the New Model Is
The orchestrator is an adapted version of GLM 5.2, post-trained specifically for the Perplexity Computer harness. In other words, rather than building a model from scratch, Perplexity took GLM 5.2 and tuned it to run the Computer environment.
According to the company, the result delivers near-frontier performance at 0.344x the cost of Opus — roughly a third of the price.
An Advisor Tool for Harder Problems
The model doesn’t operate alone. It ships with an advisor tool that natively escalates to a stronger model when needed. When the orchestrator hits a task beyond its capability, it can hand the problem up rather than failing on it — keeping the cheaper model in charge of the routine work and reserving heavier firepower for the moments that require it.
US-Hosted on Nvidia B200 GPUs
Perplexity also confirmed the infrastructure details: the model is hosted in the United States by Perplexity itself, running on Nvidia B200 GPUs.
What’s Next
The release is explicitly a research preview, and Perplexity says it will improve the model during this phase and publish full benchmarks in the coming weeks. Until those numbers arrive, the performance and cost claims come from the company’s own announcement.
FAQ
What is Perplexity’s new orchestrator model? It’s a research preview of a new model that orchestrates Perplexity Computer. It’s an adapted version of GLM 5.2, post-trained for the Computer harness.
How does it compare to Opus on cost? Perplexity says the model delivers near-frontier performance at 0.344x the cost of Opus — about a third of the price.
What happens when the model can’t handle a task? It includes an advisor tool that natively escalates to a stronger model when needed.
Where is the model hosted? It’s hosted in the U.S. by Perplexity, running on Nvidia B200 GPUs.
Are benchmarks available? Not yet. Perplexity says it will improve the model during the research preview and publish full benchmarks in the coming weeks.
