Alexandria-based TokenAI has released Forge Reasoning, an open-source dataset of 50,000 records designed to help AI models work through structured problems.
The dataset covers mathematics, science, programming, defensive cybersecurity, artificial intelligence and machine learning. Each record includes a question, step-by-step solution, final answer and a common mistake to avoid, alongside domain and topic fields.
According to the announcement, the consistent format is intended for AI training, fine-tuning, evaluation and research. The common-pitfall field gives developers an example of how a solution can fail, making the records easier to inspect and reuse in model-development workflows.
Forge Reasoning is available through Hugging Face under a CC BY 4.0 licence for TokenAI-controlled original material. The company said it has already used the dataset to train Horus Cyber Nano 1.2, an upcoming open-weights multimodal reasoning model focused on authorised security analysis, secure coding, visual inspection and defensive technical work.
TokenAI described Forge Reasoning as the first public dataset in its Forge family and said it plans to expand the collection with additional datasets, domains and evaluation resources. The nonprofit foundation was founded in 2025 by Assem Sabry.
Source: Middle East AI News
