Details
- Perplexity has released GLM 5.3 as a new model option inside Perplexity Computer, its agentic research environment.
- The model is designed for long-context workloads, supporting very large token windows suited to complex, multi-step tasks.
- GLM 5.3 is optimized for multimodal agents, indicating improved handling of mixed inputs such as text plus other formats.
- Perplexity states that GLM 5.3 outperforms the previous GLM 5.2 model on WANDR, its wide-and-deep research benchmark.
- WANDR is an evidence-backed benchmark built around large-scale research tasks, so higher GLM 5.3 scores signal better performance on realistic knowledge-work scenarios.
- The upgrade targets users running demanding agent workflows in Perplexity Computer, such as due diligence, market analysis, and large literature reviews.
- By shipping GLM 5.3 into the product stack, Perplexity continues iterating on its GLM series to push long-context, research-grade capabilities without changing the surrounding Computer interface.
Impact
By moving GLM 5.3 into Perplexity Computer and demonstrating gains on the WANDR benchmark, Perplexity strengthens its position in long-context, agentic research workloads where rivals like Anthropic and OpenAI are also investing heavily. Better performance on realistic, evidence-backed tasks should make Perplexity more attractive for professional research teams who need scalable, verifiable outputs across large document sets.
