Perplexity adds Ultrafast Computer models and Decisions v1.2

Perplexity's changelog for 19 October 2026 says Computer on the web can use Ultrafast for GPT-6 Astra and GPT-6.1 Sol. It is for Max subscribers, under Custom, then Speed. Ultrafast uses more credits than Normal or Fast.

The same note updates the Decisions API to pplx-decider v1.2. You can call a 27B standard model or a 4B Flash model, or download the open weights. Perplexity says v1.2 improves text and vision accuracy, answers multiple questions in one pass, and infers up to 5× faster. It classifies and scores text, JSON, and images as probabilities. Rubric scoring allows up to 128 questions per input. Perplexity says to validate grouped questions on your own workload, and to use one question by default for Flash. The API charges only for input tokens. The API changelog lists pplx-decider-v1.2 at $0.02 per million input tokens and pplx-decider-flash-v1.2 at $0.01, with output tokens free. Older decider ids still resolve to v1.2.

Sales-assisted Enterprise admins can choose which roles may use each connector. Access applies across Perplexity. Existing connector access stays as it is until an admin changes it. New connectors can start disabled. Enables, disables, and role changes show up in the organization's audit logs.

Who it's for: Max subscribers using Computer, developers scoring content with the Decisions API, and Enterprise admins limiting connectors.

Fits: Best for Research. Perplexity stays the first pick for a sourced answer. Ultrafast is a speed setting inside Computer, and the Decisions API is a classifier, so the rank is unchanged.