Details
- Perplexity is releasing a research preview of a new orchestrator model for Perplexity Computer.
- The model is an adapted version of GLM 5.2, post-trained specifically for the Computer harness.
- Perplexity says it reaches near-frontier performance at 0.344x the cost of Opus.
- The model includes an advisor tool that can escalate to a stronger model when needed.
- It is hosted in the U.S. by Perplexity on Nvidia B200 GPUs.
- Perplexity says the preview will keep improving and that full benchmarks will be published in the coming weeks.
- The company also says cost per task was measured against GLM 5.2 as the baseline, with one benchmark set showing roughly half the cost of Opus overall.
- On WANDR, Perplexity says GLM 5.2 plus advisor runs at 2.1x versus Opus at 6.1x.
Impact
This looks like a cost-focused move to make agentic computer use cheaper without giving up much performance, which could matter for users running many tasks at scale. By pairing a lower-cost GLM 5.2 base model with an escalation path to a stronger model, Perplexity is signaling a hybrid routing strategy that may pressure premium-only workflows in the computer-use market. The U.S.-based hosting on Nvidia B200 GPUs also keeps the deployment aligned with domestic infrastructure expectations.
