Perplexity Releases GLM 5.2 Orchestrator Model Preview
Perplexity Computer has released a research preview of a new orchestrator model. Based on a post-trained GLM 5.2 and designed specifically for the Computer harness, the model aims to deliver near-frontier performance at a significantly lower cost, powered by Nvidia hardware.
Key Details and Cost Performance
The orchestrator features a built-in advisor tool that allows it to automatically upgrade to stronger frontier models like Opus when encountering complex tasks. Perplexity measured task costs using GLM 5.2 as a baseline: in WANDR tests, the cost of GLM 5.2 with an advisor is about 2.1 times that of Opus, while using Opus alone costs 6.1 times. Overall, the new solution achieves near-frontier performance at approximately 0.344 times the cost of Opus.
Hardware Support and Future Plans
The model is currently hosted and running on Nvidia B200 GPUs in the United States. The Perplexity team stated that they will continue to add computing power to improve model quality during the ongoing research preview phase, with complete benchmarks expected to be published in the coming weeks.
2026-07-09 ~ 2026-07-11 · 7 related posts
- [source] Perplexity Releases New Orchestration Model Preview — perplexity_ai · 2026-07-09
- [source] New Model Supports Auto-Upgrades and B200 — perplexity_ai · 2026-07-09
- GLM 5.2 Orchestrator Enters Computer Harness — henrymodis · 2026-07-09
- GLM 5.2 Costs Less Than Opus — perplexity_ai · 2026-07-09
- Perplexity Launches Orchestrator Model Preview — inductionheads · 2026-07-09
- [source] Perplexity Model Hosted on B200 — nvidia · 2026-07-10
- Perplexity Unveils Its AI Orchestrator — AravSrinivas · 2026-07-11