Fractal, a Mumbai-based AI firm, has launched Fathom-R1-14B, an open-source reasoning model with 14 billion parameters.

Fathom-R1-14B is part of Fractal’s effort to establish India’s first Large Reasoning Model.

This model is designed for tasks requiring mathematical and general reasoning, according to a company post on LinkedIn.

Fathom-R1-14B is built on the Deepseek-R1-Distilled-Qwen-14B framework and features a 16K context length.

This capability allows it to process longer inputs within a single prompt.

The model also surpassed other open-source models, including o3-mini and o1-mini, in self-consistency tests that evaluate a model’s ability to provide consistent answers across multiple attempts.

🔗 Source: YourStory

🧠 Food for thought

1️⃣ The reasoning model race is redefining AI’s capabilities and applications

Fractal’s Fathom-R1-14B launch comes amid a significant shift in AI development priorities, as the industry moves from models focused on fluent text generation toward those capable of structured, logical reasoning.

This transition represents the next frontier in AI, with companies like OpenAI (O1/O3 series) and NVIDIA (Llama Nemotron family) developing specialized reasoning models that excel at mathematics, coding, and step-by-step problem solving 1.

DeepSeek-R1, which serves as the foundation for Fractal’s new model, features 671 billion total parameters with 37 billion activated for each token, showcasing the scale of investment in reasoning capabilities 2.

The performance improvements are substantial. These reasoning-focused models can show accuracy improvements of up to 20% on complex tasks compared to base models, explaining why Fractal emphasized Fathom’s performance on advanced mathematics benchmarks 3.

Industry analysts predict this competitive landscape will intensify throughout 2025, though some experts warn that innovation in reasoning models may slow as we approach fundamental scaling limits 4.

2️⃣ India’s strategic push for AI self-reliance is gaining momentum

Fractal’s development of Fathom-R1-14B as “India’s first Large Reasoning Model” represents a significant milestone in the country’s broader AI ambitions.