Monday, March 9, 2026

What happened: Perplexity launched a multi-model "Computer" system that orchestrates 19 different AI models for different tasks, moving beyond single-model approaches to task-specific model selection and execution.

Key details:

• The system routes different queries and subtasks to optimal models rather than using a single universal model
• Includes specialized models for reasoning, code generation, vision, search, and other capabilities
• Perplexity positions this as a more efficient alternative to forcing all tasks through a single frontier model
• This reflects growing understanding that no single model is optimal for all tasks
• Multi-model routing requires orchestration intelligence but can deliver better quality/cost tradeoffs
• Companies are moving away from the "one big model" paradigm toward task-specific model selection

Why it matters: The success of multi-model systems suggests the future of AI won't be dominated by single frontier models, but by orchestration layers that route work to specialized models. This distributes competitive advantage across multiple model developers and creates opportunities for systems integrators and prompt/routing engineers. It also may be more efficient and faster than always using the largest frontier model.

Practical takeaway: When building AI systems, consider whether routing different task types to specialized models (small reasoning models for simple tasks, larger models for complex reasoning) could improve both latency and cost compared to using a single powerful model for everything.