Perplexity is making one of the more tedious parts of using AI agents a little easier: deciding which model should handle a task.

The company has introduced new “effort” controls for Perplexity Computer, allowing users to tell Computer how much work a task deserves without having to manually pick a model or configure its reasoning level. The feature is rolling out on the web with four settings—Light, Standard, High, and Ultra—and Perplexity handles the model selection behind the scenes.

Perplexity Computer gets an effort slider

The new control appears as a slider in Computer’s omnibar. Instead of presenting users with a growing list of AI models and reasoning options, Perplexity reduces the decision to a simpler question: how much effort should Computer spend on this job?

At the Light setting, Computer is intended for straightforward everyday work, such as collecting vendor invoices into a spreadsheet. Standard provides a balance between reasoning depth and cost for tasks such as turning account notes and company updates into a sales brief.

High is aimed at more complicated analysis, such as weighing customer requests against engineering estimates when deciding which product work to prioritize. Ultra is reserved for open-ended problems where users want Computer to put in the maximum amount of effort, such as evaluating a new market under different growth assumptions.

The important part is that the slider doesn’t simply control how long Computer thinks. Each setting determines both the model responsible for coordinating the assignment and how deeply that model reasons.

Computer can then delegate individual pieces of the task to supporting agents, which may use models from different AI providers. In other words, users choose the desired level of effort while Perplexity takes responsibility for figuring out which combination of models should actually do the work.

That approach is particularly useful as AI agents become more complicated. A conventional chatbot might require the user to decide between several models before starting a conversation. Computer can instead treat an assignment as a workload and dynamically determine how much intelligence it needs to throw at it.

Lower effort also means lower cost

Perplexity is also tying the effort levels to cost.

The company says lower settings use less expensive models, while credit consumption ultimately depends on the work Computer performs. That gives users a simpler way to control spending without necessarily forcing them to understand the pricing and reasoning differences between individual models.

Perplexity says it evaluates models from different providers at different reasoning levels, comparing the quality of their work with their cost. When a cheaper combination produces comparable results, the company can use that combination rather than spending more on a more capable model unnecessarily.

That creates an interesting division of responsibility. Instead of asking users to constantly figure out whether a particular task deserves GPT, Claude, Gemini or another model, Perplexity wants the user to specify the importance and complexity of the job and let Computer make the technical decision.

For example, a quick spreadsheet task probably doesn’t need the same level of reasoning as a market analysis that could influence a business decision. The slider gives users a way to communicate that distinction without knowing what model should be behind it.

Manual model selection isn’t going away

Perplexity isn’t removing its existing controls for people who want more direct control.

Users who prefer to select a particular model and reasoning level can continue using the custom controls. The new effort slider is therefore an additional abstraction layer rather than a replacement for manual model selection.

That makes sense for Computer, where the whole point is to automate multi-step work. Asking users to micromanage every model decision would undermine some of the convenience an agent is supposed to provide.

The effort controls are currently available on the web. Perplexity says the feature is coming to Android and iOS soon.

Perplexity has been steadily expanding Computer beyond conventional AI chat, including local-first computing through Portable Computer for Windows and hybrid cloud/local processing on Mac. The effort slider fits into that broader shift: Computer is increasingly being positioned as an agent that decides how to execute a job rather than simply answering a prompt.

The bigger idea here is pretty straightforward: users should decide how much an AI task matters, while the AI decides which model is appropriate to get it done. As agentic systems start juggling multiple models, that could become a much more practical interface than forcing everyone to become an expert in model names, benchmarks, and reasoning modes.