Perplexity Computer Orchestrator Model (Research Preview)

A cheaper model that runs the planning layer inside Perplexity Computer, released as a research preview.

Built by post-training an adapted version of GLM 5.2 specifically for the Perplexity Computer harness
Runs at 0.344x the cost of Opus while staying near frontier performance
Carries an advisor tool that escalates to a stronger model on its own when a step needs one
On the WANDR benchmark, GLM 5.2 plus advisor runs at 2.1x the baseline cost against 6.1x for Opus, averaging roughly half the cost of Opus
Hosted in the United States by Perplexity on Nvidia B200 GPUs