Breaking

Perplexity Debuts GLM-5.2-Based Orchestrator at 0.344x the Cost of Opus 4.8

July 9, 2026 at 10:57 EDT

Foundation Models
AI Agents
Software Dev & Coding

On July 9, 2026, Perplexity released a research preview of a new orchestrator model for its agentic platform, Perplexity Computer, built on Z.ai (Zhipu AI)'s GLM-5.2. The company says it delivers performance on par with Claude Opus 4.8 at 0.344x the cost.

July 9, 2026 · Perplexity

Opus-Level Performance at One-Third the Cost

Perplexity's new orchestrator for its agentic platform — built on Z.ai's open-weight GLM-5.2 — claims to match Claude Opus 4.8 while running at just 0.344× the cost.

0.344×

Cost relative to Opus 4.8

Token context window

~40B

Active params (of ~750B total, MoE)

Cost on the WANDR benchmark

Lower is cheaper — GLM-5.2 with its advisor tool runs at roughly one-third of Opus.

2.1×

GLM-5.2
+ advisor

6.1×

Opus 4.8

Coding benchmarks: close but behind

Terminal-Bench 2.1

GLM-5.2

81.0

Opus 4.8

85.0

SWE-bench Pro

GLM-5.2

62.1

Opus 4.8

69.2

Developers praise

Strong cost-performance for coding
Fast response on long tasks
Handles complex, conflicting prompts

Caveats

Weaker on general, non-coding queries
Perplexity Computer version still in preview
Broad evaluations yet to accumulate

How the orchestrator escalates

GLM-5.2 orchestrator

→

Advisor tool detects complex task

→

Escalate to more powerful model

Post-trained for the Perplexity Computer harness · hosted on NVIDIA B200 GPUs in the U.S. · GLM-5.2 is open-weight under the MIT license.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.