Fractal rolls out Fathom-R1-14B, an open-source AI model for reasoning

AI unicorn Fractal has launched Fathom-R1-14B, an open-source reasoning model with 14 billion parameters, designed to handle mathematical and general reasoning tasks. 

Announcing the development in a LinkedIn post, the newly launched model was developed by the Mumbai-based firm’s AI Research team at a post-training cost of $499. 

The AI model, built on top of Deepseek-R1-Distilled-Qwen-14B, supports a 16K context length, meaning it can process and understand longer inputs in a single prompt. It also demonstrated strong performance on advanced math benchmarks like AIME (American Invitational Mathematics Examination) and HMMT (Harvard-MIT Mathematics Tournament). 

Fathom-R1-14B outperforms other open-source models such as o3-mini and o1-mini, and exceeds o4-mini-low in self-consistency tests, which measure how reliably a model gives the same answer across multiple runs.

Also Read

Sarvam AI brings 24B-parameter LLM for Indian languages, reasoning

The model was developed as part of Fractal’s proposal to build India’s first Large Reasoning Model (LRM) under the IndiaAI Mission.

“Today’s large pre-trained AI models are great at summarization, information retrieval, and content generation. But the next frontier is building systems that can think, reason, pla,n and act. It’s about building systems that can work with pre-trained models - and accomplish complex real-world tasks through better reasoning,” said Srikanth Velamakanni, Co-founder of Fractal, in an earlier LinkedIn post.

“That’s why we, at Fractal, proposed building India’s first Large Reasoning Model (LRM) - a next-generation AI system built to work with open-source LLMs, trained on Indian data, and designed to tackle real-world complexity. It’s a bold step, but a necessary one,” he added.  

Prior to this, the company last year launched Vaidya.ai, a multimodal AI platform to provide offer free and accessible healthcare assistance to users.