Perplexity

New research: We developed JOLT, a post-training method where one model acts as both teacher and student, with the teacher trained to give feedback the student can use. On MATH, it matched GRPO's accuracy with 6.5x fewer training tokens. perplexity.ai/hub/blog/a-goo…

Reinforcement learning methods like GRPO give one reward per attempt, which doesn't show which steps helped. Self-distillation adds per-token feedback from a teacher. The same model plays the teacher, with extra information like a verified solution, and the student, without it.

Perplexity Computer can now run fully local on your Mac. Work with your native Mac apps and local files, and run tasks without sending sensitive data to the cloud. The model and the agent both run on your Mac. Available now in the Perplexity Mac app.

Vídeo

Local tasks use PPLX Pro, our post-trained version of Qwen 3.8, and don't consume Computer credits. For more complex work, Hybrid can split the task between a cloud models and a local model on your Mac, keeping sensitive data private. Read more: perplexity.ai/hub/blog/perpl…

Gemini 4 Argon sets a new record on WANDR. It is now available in Perplexity for Pro and Max subscribers.

NVIDIA’s Nemotron 3.5 Super VL is now available in Perplexity for Pro and Max subscribers, and in Agent API.

Vídeo

We're open sourcing SPACE, the sandbox platform that runs every task in Perplexity Computer. SPACE gives each agent task its own Firecracker microVM and controls credentials and outbound traffic from outside the VM. Available now under Apache 2.0: github.com/perplexityai/S…

Agents in Computer run commands, install packages, and call external services with root access inside the sandbox. The host stays outside the guest's control, and a per-sandbox gateway enforces egress policy on every outbound connection.

Ultrafast versions of GPT-6 Astra and SOL 6.1 are now available in Perplexity Computer for Max subscribers. Both generate tokens faster, completing long Computer tasks sooner. Choose Astra or SOL 6.1 under Custom, then set speed to Ultrafast.

Vídeo

Our sandbox, SPACE, handles millions of executions a day with an error rate below 0.0002%. We’re sharing how it uses cryptographic identity to keep requests tied to the right execution as sandboxes pause, resume, and move. Read more: perplexity.ai/hub/blog/how-s…

When a sandbox resumes on another machine, it keeps its sandbox ID but starts a new execution. A stale route can send a valid request to the older execution and return old data. Checking the sandbox ID alone won’t catch the difference.

This is how we built the filesystem behind Perplexity Projects as a Git forge for agents. Each agent works in its own sandbox and commits to a shared repository where people and other agents can build on its work across sessions. Read more: perplexity.ai/hub/blog/build…

Agents collaborating on a long-running project face the same problem as engineers sharing a codebase. Work must outlive ephemeral sandboxes, concurrent writes must not overwrite each other, and history must show who changed what.

We're releasing pplx-embed-v2-late, two late-interaction embedding models that retrieve text, images, and pages with a shared embedding space for cross-model querying. Both models achieve frontier performance and are publicly available on Hugging Face. perplexity.ai/hub/blog/multi…

Dense embedding models compress each document into one vector, which loses detail as pages get longer or more visual. pplx-embed-v2-late keeps a 128-dimensional vector per token and scores with MaxSim, so each query token is matched to its closest token in the document.

Perplexity Computer now creates interactive charts and visualizations directly in your thread. For financial data, Computer uses @tradingview Lightweight Charts for candlesticks, volume, and moving averages.

Vídeo

We built a new way to train contextual embedding models, which encode each chunk of a document with the whole document in view. pplx-embed-v2-context-9b-preview sets a new state of the art on ConTEB and @turbopuffer's new, privately held context-bench. perplexity.ai/hub/blog/conte…

Retrieval systems often split long documents into chunks, but this strips away the surrounding context. Contextual embedding models fix this by encoding the whole document once and pooling chunk vectors afterward. They are usually trained on one gold chunk per query.

Introducing Automations in Perplexity Computer. Automations are for ongoing work. They can take action in response to event-based triggers or on a schedule. Automations work with your memory, skills, and connected apps like Slack, Gmail, Outlook, and Linear.

Vídeo

Read more about Automations in Perplexity Computer: perplexity.ai/hub/blog/compu…

Agent governance is an engineering problem. We’ve built safeguards into Perplexity’s infrastructure, harnesses, and tools, and put them to work across our products. Today we're sharing how we engineer safer agents: perplexity.ai/hub/blog/how-w…

SPACE runs every Computer task in an isolated microVM, with platform credentials and network controls outside the sandbox. Snapshots let us roll back changes. Enterprise customers can revoke encryption keys to make protected data unreadable; the agent can’t override it.

We’re partnering with Nvidia and 100+ industry partners to build infrastructure that contains rogue AI agents. In this research, we gave 9 AI models root access inside SPACE and told them to break out. Across 108 runs, none breached the VM boundary. perplexity.ai/hub/blog/escap…

In the recent OpenAI-Hugging Face incident, agents compromised OpenAI's internal research infrastructure as well as Hugging Face's. SPACE is built to contain exactly that, whether an agent fails by accident or tries to escape.

Introducing Fast Search in the Perplexity Search API. Fast Search runs on Photon, our new Rust-based retrieval and ranking service that we built with a small team of engineers and hundreds of agents. It returns 95% of search results in 230 ms or less.

@perplexitydevs Across the six agent benchmarks, it cut cost per task by 68% compared with the default preset, with comparable quality. At 160 ms, Fast Search had the lowest median latency of any search API we compared.

Portable Computer for Windows is now available on @AMD Ryzen AI Max Series processors. It makes it easy to run local AI agents that work with your connected apps and local files. Kick off tasks or schedule recurring work that runs entirely on your device.

Vídeo

Portable Computer handles tasks locally, running the agent harness and model on your device, and asks your permission when it needs help from a cloud model.

New research: We post-trained a Computer model to learn from its own errors using hint-guided self-distillation. In a live A/B test, a later trained checkpoint reduced tool-call failures by 21.2% relative to an earlier checkpoint.

After reinforcement learning in synthetic environments, we train on real-world sessions that expose failures beyond our designed scenarios. Our sampling pipeline excludes sessions with personally identifiable information (PII) and sessions from users who opted out of training.

GPT-6 Sol is now available in Perplexity and Computer. It's now the default Light option in Computer's effort selector.

Claude Opus 5.5 is now available as the Standard effort level in Perplexity Computer. On our 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 Computer can now create videos using MiniMax H3 and ByteDance Seedance 2.5. Ask Computer for a campaign clip, product demo, or social asset, and it produces finished video next to copy and creative in same thread. Available now for Pro and Max subscribers.

Vídeo

We’re rolling out effort controls in Computer’s model selector. 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.

Vídeo

For more control, you can customize model selection, reasoning level, and available speed settings.

We’re publishing research on how we built CobbleDB, our key-value database that serves web content for Perplexity search. Two engineers and a team of hundreds of proactive, always-on AI agents built the core infrastructure in two months.

We accelerated CobbleDB development with a team of proactive coding agents. The agents provided continuous inspection and follow-through across sessions under the direction of engineers responsible for architecture, review, and production authorization.

Perplexity Computer will now come pre-installed on HP ZBook Ultra G3a. . Use Computer to run complex multi-step work from a simple interface. Computer agents are grounded in accurate deep research and connected to hundreds of tools, now including Autodesk.

@autodesk We’ve worked with HP and Autodesk to connect Computer to Revit through Autodesk’s MCP server. Computer can read building data from your open Revit model to write reports, export views and schedules, and draft requests for information.

Portable Computer is now available on Windows PCs with @NVIDIA RTX GPUs. Run the harness, agents, and models locally on your PC. Work with local files and connected apps without sending tasks to the cloud. Use frontier cloud models when needed.

Vídeo

We’ve also added support for local MCP and scheduled tasks. Connect Portable Computer to your own tools and app integrations through local MCP. Give Computer recurring work and schedule it to run on your PC while you’re away.

We evaluated 13 retrieval models across three relevance sets, using Recall@1000 as the primary metric. pplx-embed-v1-4b leads Web Ranking (65.73) and Combined (69.11), while Nemotron-3-Embed-8B leads Citation (61.68).