Perplexity ships a GLM 5.2-based orchestrator that matches frontier scores at one-third the cost
rohanpaul_ai · x · 2026-07-23
Perplexity says its new computer orchestrator is an adapted GLM 5.2 post-trained for the Computer harness.
- It reportedly reaches near-frontier performance while costing about 0.344x of Opus, or roughly one-third the cost.
- On Terminal-Bench 2.1, the setup scores 80, ahead of Opus 4.8 (76) and GPT-5.5 (78).
- On WANDR, the advisor build scores 29 at 2.1x baseline cost, versus Opus 4.8 at 30 but 6.1x cost.
- Perplexity says the model runs in the U.S. on Nvidia B200 GPUs and uses an internal advisor tool to escalate to stronger models only when needed.
More from Companies & People
- Post says the U.S. led open models first, then let closed models take over — cocktailpeanut · 2026-07-23
- For enterprise AI agents, the first question is ownership and governance, not Copilot Studio vs. Foundry — WirelessLife · 2026-07-23
- OpenAI raises projected compute spend to $750B and starts designing its own data centers — kimmonismus · 2026-07-23
- WAIC speaker says world models are gaining real momentum in China — smtabatabaie · 2026-07-23
- User criticizes Anthropic strategy: RLHF unbearable, Chinese models more practical — StewartalsopIII · 2026-07-23
- AI-culture joke says you now have to LARP to succeed — KadriJibraan · 2026-07-23