Perplexity began rolling out effort controls in its Computer agent on 17th September 2026, putting model choice and reasoning depth behind a single four-step slider. Lower settings run on less expensive models, and credit usage still depends on the work Computer performs.
The Big Picture
- Perplexity rolled out effort controls in its Computer agent, letting a slider set both the model and how deeply it reasons.
- The slider carries four settings. Light, Standard, High, and Ultra each map to a different model and reasoning level.
- Lower settings use less expensive models, so what a user spends tracks the setting picked and the work performed.
- Custom controls remain for people who want to name a specific model and set its reasoning level themselves.
- The controls are live on web only. Perplexity says Android and iOS support is coming but has not given a date.
Perplexity hands model choice to a slider
Perplexity placed the control in Computer’s omnibar, where a slider moves across four positions. The company selects the model and the reasoning level sitting behind whichever position a user lands on, according to its announcement.
Perplexity framed the change as a direct answer to what users asked for. “People told us they wanted to decide how much effort a task deserved and what they were willing to spend, while leaving model selection to Perplexity,” the company said.
That trade sits at the center of the feature. A user gains a spend dial and gives up the ability to know which model ran the job, unless they switch back to custom controls.
The slider also lags the product it governs. Computer has been available on iOS since March 2026, and Perplexity shipped a macOS app carrying its Personal Computer assistant in May 2026. The effort controls launched on web alone.
We’re rolling out effort controls in Computer’s model selector.
— Perplexity (@perplexity_ai) September 17, 2026
Effort presets combine the orchestrator model and reasoning depth to control how deeply and efficiently Computer works through a task.
Available now on web. Coming to mobile and desktop soon. pic.twitter.com/7Kuj6vavcf
What each setting is for?
Perplexity attached a worked example to every step on the slider:
- Light: straightforward, everyday tasks, such as collecting vendor invoices in a spreadsheet.
- Standard: balanced reasoning and cost, such as combining account notes and company updates into a sales brief.
- High: complex analysis, such as weighing customer requests against engineering estimates to prioritize product work.
- Ultra: maximum effort on open-ended problems, such as evaluating a new market under different growth assumptions.
The cost line is the one to read twice. Lower settings use less expensive models, Perplexity said, and credit usage depends on the work Computer performs. The announcement published no per-setting credit rate and named no model behind any step, so the slider communicates relative spend without attaching a number to it.
The orchestrator picks, then delegates
An effort setting selects the model that coordinates an assignment and decides how deeply that model reasons. The orchestrator then works out an approach and hands parts of the job to supporting agents, and those agents can run on models from different providers. One assignment can therefore cross vendors without the user choosing any of them.
Perplexity says its experience answering billions of queries informs the matching, and that it tests models from multiple providers at different reasoning levels, weighing quality of output against cost. When a cheaper combination produces comparable results, the company routes the work there and leaves the user more budget for harder assignments.
Perplexity published no benchmark data alongside the announcement, so the claim that cheaper combinations match pricier ones rests on internal testing nobody outside the company can reproduce. Those routing economics matter to a business that raised against a $20 billion valuation and sells Computer on flat monthly tiers, as Perplexity’s funding history shows.
What Perplexity has not said
The announcement leaves four questions open:
- Which models sit behind Light, Standard, High, and Ultra, and can Perplexity change that mapping without telling users?
- How many credits does each setting consume on the same assignment?
- When do Android and iOS get the slider?
- Does the chosen effort level apply to the supporting agents an orchestrator delegates to, or only to the orchestrator?
Users already running Computer should check the slider position in the omnibar before starting a long assignment, since that position now decides the model. Naming a model directly in custom controls helps reduce surprise credit usage. It does not prevent it, because spend still follows the work the agent does.
Why It Matters?
This release turns a model picker into a spend dial. The useful part is that one control now sets two things that used to move separately, which is closer to how people budget a task in the first place. The cost is visibility. A user who picks Ultra is trusting Perplexity’s routing rather than a named model, and the announcement offers no way to audit that choice after a job finishes.
What’s next is the mobile rollout, which Perplexity has promised without a date. Watch for two things after that: a published credit rate per setting, and any disclosure of which models sit behind each step. Both decide whether the slider is a real budget tool or a label. Teams that already pay for agents such as Perplexity’s email assistant will also want to know whether an administrator can hold a workspace to one setting.
