OpenRouter has integrated Perplexity AI's open-weights multimodal decision model, Perplexity Decider v1.1 (pplx-decider-v1.1-27b). The model is priced at $0.02 per million input tokens with free output.
Perplexity has released pplx-embed-v2-late, two late-interaction embedding models that retrieve text, images, and pages within a shared embedding space, now publicly available on Hugging Face.
Perplexity Developers has made pplx-decider-v1.1-27b available as an updated open-weights multimodal decision model. Priced at $0.02 per million input tokens, the model costs half as much as v1 and ranks highest on the Hugging Face Decision Index 0.3 benchmark according to the team.
Aravind Srinivas says Perplexity is starting to deploy Perplexity Computer on Vera, which he calls far better than x86. The stated goal is to own the sandboxes and optimize them for that silicon.
Perplexity is introducing Automations in Perplexity Computer for ongoing work. They can act on event-based triggers or a schedule, and work with memory, skills, and apps including Slack, Gmail, Outlook, and Linear.
Perplexity says its Secure Intelligence Institute works with researchers at Stanford, CMU, Duke, Columbia, Ohio State, and UVA, and is also working with NVIDIA and the Open Secure AI Alliance to share tools and findings.
Perplexity says Bumblebee checks developer machines for risky packages and extensions, while Computer proposes better detection rules from Bumblebee and Numbat findings. Humans approve every change before it ships, and agents cannot approve their own changes.
Perplexity says it has built safeguards into its infrastructure, harnesses, and tools and applied them across its products, and has published an account of that approach.
Perplexity says it evaluated 10 third-party sandbox platforms and found that 8 were vulnerable to similar network-policy bypasses. It disclosed the findings to the affected vendors and received a response from each.
Perplexity says Claude Opus 5.0 identified an IP-sharing route in its SPACE evaluation and declined to use it. Fable and GPT-6 Astra refused the tasks outright.
Perplexity is partnering with Nvidia and more than 100 industry partners on infrastructure meant to contain rogue AI agents. In its SPACE research, nine models given root access did not breach the VM boundary across 108 runs.
Perplexity’s Portable Computer runs in its Windows app and is available to all Consumer and Enterprise Pro and Max subscribers.
Perplexity says Portable Computer works across connected apps, including Slack, Gmail, and Google Drive, as well as files on a device. Tasks that run entirely locally keep files, data, and answers on the device and do not use Computer credits.
Perplexity’s Portable Computer for Windows now runs on AMD Ryzen AI Max Series processors, with local AI agents that use connected apps and local files and can run scheduled work entirely on the device.
Perplexity post-trained a Computer model to learn from its own errors using hint-guided self-distillation. It said the 21.2% reduction was relative to an earlier checkpoint.
GPT-6 Sol can be used in Perplexity and Computer, where it is the default Light option in the effort selector.
Perplexity has made Claude Opus 5.5 available as the Standard effort level in Perplexity Computer. On Perplexity’s WANDR benchmark, it scored 0.610 at $4.13 per task, slightly outperforming Claude Fable 5.1 while costing 67.6% less per task.
Perplexity said Perplexity Computer can now create videos using MiniMax H3 and ByteDance Seedance 2.5, producing finished video next to copy and creative in the same thread. It said the capability is available now for Pro and Max subscribers.
