Details
- Perplexity announces that the DeepSeek V4 Pro model, hosted in the United States, is now available within Perplexity Computer.
- The company evaluated DeepSeek V4 Pro on its WANDR wide-and-deep research benchmark, which measures performance on complex, evidence-heavy tasks.
- DeepSeek V4 Pro scored 0.359 on WANDR at a price of $0.75 per task, indicating strong efficiency on structured research workloads.
- Perplexity states that this cost is 62% cheaper than the next model on the benchmark’s cost-performance frontier, underscoring favorable economics for users.
- Adding DeepSeek V4 Pro broadens the set of models available in Perplexity Computer, giving customers a lower-cost option for high-volume research and knowledge work.
- The U.S. hosting detail suggests a deployment tailored to customers with data residency, compliance, or latency requirements in North America.
- Positioning DeepSeek V4 Pro on the WANDR cost-performance frontier signals Perplexity’s focus on transparent, benchmarked trade-offs rather than purely raw model scores.
Impact
By bringing DeepSeek V4 Pro into Perplexity Computer with clear WANDR cost-performance data, Perplexity strengthens its positioning in agentic research workloads while appealing to customers who need U.S.-hosted models and predictable per-task economics. This move pressures rival AI platforms to disclose similarly granular benchmark and cost metrics, shifting competition toward measured efficiency rather than headline capabilities.
