Perplexity Computer Adds Ultrafast Inference
Perplexity Computer adds ultrafast inference for GPT-6 Astra and SOL 6.1 on its Max tier, alongside a warning about rapid credit consumption.
The speed setting cuts latency for GPT-6 Astra and SOL 6.1 in autonomous workflows, but drains usage credits rapidly and requires a Max subscription.
Perplexity rolled out an ultrafast inference option on Wednesday (October 14, 2026) for OpenAI’s frontier models inside Perplexity Computer. The speed setting, officially named “Ultrafast”, is available for both GPT-6 Astra and SOL 6.1. The announcement came directly from the company and CEO Aravind Srinivas, who described the output as “super fast” while pairing the news with an explicit “credit drain warning” due to the rapid consumption of account credits.
Perplexity Computer serves as the company’s dedicated environment for autonomous agent workflows. Inside this workspace, the system plans tasks, searches the web, writes code, and runs tests in isolated sandboxes without manual prompts at every step. In long workflows that require dozens of sequential model calls, generation speed determines whether an automated run finishes in five minutes or drags on for close to an hour.
How Ultrafast Inference Works
To activate the setting, users open the Perplexity Computer interface, head to the “Custom” menu, select either GPT-6 Astra or SOL 6.1, and toggle the speed switch to “Ultrafast”. This configuration bypasses standard serving latency in favor of a low-latency pipeline designed to keep agents moving between tool calls without idle pauses.
Perplexity did not publish formal benchmark data alongside the release. There are no official throughput figures in tokens per second (the basic fragments of words language models generate) and no side-by-side charts measuring the speedup against standard inference tiers. The acceleration is noticeable in the interface, but the actual time saved depends on the prompt structure and the complexity of the task.
Credit Consumption in Long Workflows
That speed boost carries an immediate cost on the user’s account. In his warning, Srinivas noted that Perplexity usage credits drain noticeably faster while Ultrafast mode is active. Advanced reasoning models require heavy compute; running those operations at high speed concentrates resource usage into a narrow time window.
The company has not disclosed the exact credit multiplier applied to Ultrafast runs or published a tiered pricing breakdown. Srinivas’s note acts as an informal caution so subscribers avoid burning through their monthly allotments across a small number of automated jobs.
Max Tier Exclusivity and Access Limits
The Ultrafast option is not open to free users or Perplexity Pro subscribers. Access is restricted to Perplexity Max, the service’s most expensive tier tailored to developers and heavy workloads. Perplexity has not stated whether the feature will eventually reach lower subscription plans or appear on its external developer API.
The release builds on an established technical relationship between both companies. Perplexity previously deployed GPT-6 Astra inside its own engineering pipelines for automated testing and code monitoring, according to an OpenAI case study. Offering accelerated Astra and SOL 6.1 setups inside Perplexity Computer shortens wait times for developers running agentic tasks, even if the price of that speed keeps it confined to a high-end plan.