Fractal has been selected by the Government of India—under the flagship IndiaAI Mission—to develop the country’s first Large Reasoning Model, a landmark step that elevates India’s position in the global AI landscape. At the forefront of this initiative is Fathom DeepResearch 4B, an advanced agentic AI system engineered to perform long-horizon, PhD-grade inquiry. Purpose-built for autonomous exploration, discovery, and synthesis across the open web, the model represents a pivotal advancement in India’s capability to produce verifiable, high-quality AI-driven research at scale.
Fathom DeepResearch 4B stands apart through its multi-agent self-play framework, which enables the system to generate sophisticated datasets tailored for complex question answering. This architecture not only enhances the breadth and depth of its reasoning abilities but also ensures continuous improvement as agents challenge and refine one another. The system integrates innovations like RAPO, a mechanism that stabilises reinforcement learning during extended reasoning cycles, helping the model maintain coherence and accuracy even in multi-step problem-solving scenarios. Complementing this is a Steerable Step-Level Reward system that guides the model toward safe, reliable, and contextually appropriate tool usage—an increasingly crucial requirement as agentic systems interact more dynamically with external information sources.
A defining capability of DeepResearch 4B is its structured Report Synthesis Protocol, which transforms raw research traces into organised, citation-backed, publication-ready outputs. This process ensures that the knowledge the system uncovers is not only comprehensive but also verifiable, transparent, and suited for academic, institutional, or enterprise-grade consumption. It elevates the quality of automated research from simple summarisation to disciplined knowledge construction.
By launching DeepResearch 4B and taking on the responsibility of building the nation’s first Large Reasoning Model, Fractal is laying the foundation for a new generation of AI research infrastructure in India. The initiative promises a future where scalable, rigorous, and high-impact research can be produced far faster than traditional methods allow—empowering government bodies, academic institutions, and enterprises to unlock new discoveries, innovate rapidly, and compete globally in the era of autonomous intelligence.
