Perplexity Releases GLM 5.2-Based Orchestrator Model for Computer at Third the Cost of Opus.

Perplexity announced on LinkedIn a research preview of a new orchestrator model for Perplexity Computer, adapted from GLM 5.2 and post-trained specifically for the Computer harness. The company claims the model delivers near-frontier performance at 0.344x the cost of Anthropic's Opus, a significant cost reduction for users running complex AI workflows.

The new model is hosted in the U.S. by Perplexity on Nvidia B200 GPUs and includes a built-in advisor tool that can automatically escalate tasks to a stronger model when needed. This dynamic routing capability could help balance cost efficiency with performance for Computer Pro and Max subscribers.

Perplexity's claim of 0.344x the cost of Opus positions the GLM 5.2-based model as a budget-friendly alternative for users who need reliable AI orchestration without the premium pricing of frontier models. The company has not yet released detailed benchmarks, but promises to publish full performance metrics in the coming weeks as the research preview matures.

The orchestrator model sits at the center of Perplexity Computer's workflow engine, determining which specialized models to route tasks to and coordinating multi-step operations. By using a cost-optimized base model that can escalate to more powerful options when necessary, Perplexity aims to reduce the average credit spend per task while maintaining quality output.

This follows recent additions of Grok 4.5 and GPT-5.6 models to the platform, showing Perplexity's strategy of offering multiple orchestrator options at different price and performance tiers.

The advisor tool represents a key technical feature. Rather than requiring users to manually select more powerful models for complex queries, the orchestrator can recognize when a task exceeds its capabilities and automatically hand off to a stronger model. This creates a two-tier system: routine operations run on the cost-efficient GLM 5.2 adaptation, while edge cases that demand higher reasoning ability trigger escalation.

Perplexity has been iterating rapidly on orchestrator models over the past months, previously offering Claude Fable 5 and other options. The company also launched Computer Analytics to help users track credit consumption across different models, addressing a common pain point as teams manage AI budgets.

By labeling this release as a research preview, Perplexity signals that the model will continue to evolve based on user feedback and internal testing. The company plans to improve the model during the preview period before potentially promoting it to general availability. Users should expect refinements to the advisor tool's escalation logic, performance optimizations, and expanded benchmark coverage.

The use of Nvidia's B200 GPUs, announced earlier this year as part of the Blackwell architecture, suggests Perplexity is investing in cutting-edge infrastructure to support both cost efficiency and low-latency inference. The U.S.-based hosting may also appeal to enterprise customers with data residency requirements.

Perplexity has not specified whether the GLM 5.2-based orchestrator will replace existing options or join the rotation of available models for Computer users. Given the platform's history of offering multiple orchestrator choices, it seems likely this will become another option in the model selection menu rather than a wholesale replacement.