Perplexity Computer Launches AI Orchestra: 19 Models Collaborate to Replace $30K Financial Terminals
A New Era of Autonomous Computing
Today, Perplexity unveiled Perplexity Computer — a groundbreaking multi-agent system that orchestrates up to 19 state-of-the-art AI models in parallel to execute end-to-end tasks: research, design, coding, deployment, and project management — all without human intervention.
How It Works
- Dynamic Model Assignment: Claude Opus acts as the conductor — intelligently routing subtasks to optimal specialists (e.g., Gemini for research, Grok for speed, Claude for reasoning).
- Autonomous Execution: Runs unattended for hours or days in the background.
- Built-in Infrastructure: Persistent memory, native file & internet access, and hundreds of pre-integrated connectors.
Real-World Impact: From Concept to Production
📈 Financial Terminal Disruption
Perplexity Computer built a Bloomberg Terminal–level financial analytics interface — complete with live charts, company profiles, financial summaries, and market insights — in seconds, not months.
💡 Bloomberg Terminal costs $30,000/year. Perplexity Computer delivers comparable capabilities at scale — with flexible, usage-based pricing.
🌍 Satellite Tracking Web App — In One Prompt
With a single instruction, it:
– Pulled TLE orbital data for ~1,500 satellites from CelesTrak,
– Ran SGP4 orbit propagation using satellite.js,
– Rendered real-time global satellite positions with sub-second updates.
💻 Developer Workflow Acceleration
- Assigned exploration to Gemini, code generation to Claude Opus, and PR submission to GitHub — autonomously.
- Built detailed NVIDIA valuation models from zero-shot analysis.
Behind the Scenes: The Architect & Vision
Led by Alex Graveley, former GitHub Copilot Chief Architect and creator of the AI-native browser Comet, Perplexity Computer embodies a radical redefinition of computing.
Founder Aravind Srinivas’s seminal essay declares:
“AI is the computer.”
He introduces the internal ‘ASI’ project — evolved from a Slack plugin into a full digital employee with:
– Task decomposition & sub-agent creation,
– Tool orchestration across 100+ APIs,
– Cloud sandbox execution,
– Native filesystem, browser, and shell permissions.
Srinivas argues that monolithic LLMs are productivity bottlenecks — true intelligence emerges from heterogeneous model collaboration, where each agent plays to its strength: code → coder models, logic → reasoners, speed → lightweight inference engines.
🔑 The new moat? Not raw compute — but adaptive orchestration across diverse AI families.
Anthropic’s Parallel Leap: Human-Like Computer Use
In tandem, Anthropic acquired Vercept, an AI startup founded by ex-Meta/Ai2 researchers (Ross Girshick, Luca Weihs, Kiana Ehsani), focused on enabling AI to see, understand, and act within GUI software — just like humans.
- Vercept’s tech bridges the “perception-action gap” in agent workflows.
- Integrated into upcoming Claude versions, it boosts OS-level interaction capability.
- On the OSWorld benchmark, Claude Sonnet 4.6 jumped from <15% to 72.5% success rate, nearing human performance.
Availability & Access
- ✅ Web Access: Available now to Perplexity Max subscribers (10,000 tokens/month included).
- 💰 Pricing: Usage-based — choose specific models per subtask and set token limits.
- 🎁 Launch Bonus: All users (new & existing) receive 20,000 bonus tokens valid for 30 days.
Sources: Perplexity Computer Official Page, X (Nat Jin), Perplexity Blog, Anthropic Announcement
Published by AITNT — The Leading AI News Platform