Perplexity Adds DeepSeek V4 Pro to U.S.-Hosted Computer Workflows.

Perplexity has added DeepSeek V4 Pro to Perplexity Computer, giving its agentic workspace another model option for completing multi-step tasks. The company says the model is hosted in the United States and delivered notably lower benchmark costs than its closest competitor on a cost-performance basis.

In an August 18 LinkedIn announcement, Perplexity reported that DeepSeek V4 Pro scored 0.359 on WANDR at a cost of $0.75 per task. According to the company, that makes it 62 percent cheaper than the next model on the benchmark’s cost-performance frontier.

The addition is significant because Perplexity is positioning Computer as a system that can match different models to broader workflows, rather than relying on one default model for every job. DeepSeek V4 Pro expands that pool with an option focused, at least according to Perplexity’s published evaluation, on balancing task performance and inference cost.

A cost-performance frontier compares models that provide the strongest measured result at a given cost. A model sits on that frontier when another tested option cannot deliver both better performance and a lower price under the same evaluation conditions.

Perplexity’s reported figures therefore suggest that DeepSeek V4 Pro is economically competitive for the tasks included in WANDR. The 62 percent figure does not mean it is 62 percent cheaper than every other model, or that users will see an equivalent reduction in their subscription costs. It refers specifically to the next model on the cost-performance frontier in Perplexity’s evaluation.

The distinction matters. Benchmark costs can reflect model inference, task execution or an evaluation-specific accounting method. Perplexity did not publish additional methodology in the announcement, such as the number and type of tasks tested, the models included in the comparison, or whether repeated runs were averaged. It also did not provide an independent evaluation of the 0.359 score.

For small businesses and builders, the practical takeaway is that Perplexity believes the model can complete certain computer-based tasks at a comparatively low operating cost. That could become useful for workflows involving repeated research, analysis or other multi-step jobs, particularly when they need to run at volume. Actual results will still depend on the task, instructions and tools available to the model.

Perplexity explicitly highlighted that DeepSeek V4 Pro is hosted in the U.S. That may be relevant to organizations evaluating where AI workloads are processed, especially given the scrutiny surrounding models developed by Chinese AI companies.

U.S. hosting, however, should not automatically be read as a complete data residency or compliance guarantee. The brief announcement did not detail where all associated data is stored, how long prompts and outputs are retained, or which security and privacy controls apply. Businesses with regulated or sensitive information will still need to consult Perplexity’s formal terms and security documentation before deploying the model in production workflows.

The launch also fits a broader pattern of Perplexity adding models to Computer rather than treating its agent product as a single-model environment. It follows the company’s addition of Grok 4.6 across Answer and Computer workflows and its deployment of GPT 5.6 Terra and Luna across Computer workflows.

That widening selection could make model economics increasingly important. The most capable model is not necessarily the best choice for every routine task, particularly when an agent must make several calls to finish one workflow. Lower per-task costs can compound quickly at scale.

Perplexity has not disclosed which subscription tiers include DeepSeek V4 Pro, whether usage limits apply, or whether the model is available in every region. It also did not provide technical details such as context-window size or supported tool capabilities. For now, the clearest claims are narrow: DeepSeek V4 Pro is available in Perplexity Computer, it is hosted in the U.S., and it delivered the company’s stated WANDR result at $0.75 per task.

What is Perplexity DeepSeek V4 Pro?

DeepSeek V4 Pro is a model now available inside Perplexity Computer for carrying out computer-based, multi-step tasks. Perplexity says its deployment is hosted in the United States and scored 0.359 on its WANDR evaluation.

How much does DeepSeek V4 Pro cost in Perplexity Computer?

Perplexity reported an evaluation cost of $0.75 per task, but that is a benchmark figure rather than announced customer pricing. The company has not specified which subscription plans include the model or whether separate usage limits and charges apply.

Why does the WANDR result matter?

The result is intended to show the model’s balance between measured performance and execution cost. Perplexity says DeepSeek V4 Pro was 62 percent cheaper than the next model on the cost-performance frontier, although it did not publish enough methodology in the announcement to support a broader comparison across all workloads.