The Next Breakthrough In AI Agents Is Here

TL;DR

Manas is a multi-agent AI system that uses a planner agent, subagents, and 29 integrated tools to complete real-world tasks like travel planning and financial analysis. It scores 86.5% on the GAIA benchmark (vs. 74% for OpenAI’s Deep Research) and costs ~$2 per task.

Key insights

• Manas replaces a single neural network with a multi-agent system: a planner agent creates a master plan, subagents (each with a delineated domain) execute subtasks, and an executor agent synthesizes t
• It uses a proprietary “chain of thought injection” technique to maintain stability after dozens of reasoning and tool-use rounds.
• Manas scores 86.5% on GAIA (human baseline ~92%), outperforming OpenAI’s Deep Research (74%).