S1-Base: Scientific Foundation Model

S1-Base is a series of scientific foundation models developed by ScienceOne-AI, designed for scientific reasoning and complex disciplinary tasks. The S1-Base-8B model, based on Qwen3-8B, is trained on an extensive corpus of 170 million scientific papers and fine-tuned with millions of high-quality scientific reasoning data.

Key Capabilities

• Scientific Reasoning: Systematically learns core theories, laws, and specialized knowledge across six foundational disciplines: mathematics, physics, chemistry, earth science, astronomy, and biology.
• Multidisciplinary Expertise: Strengthens subject capabilities through a curriculum-based training strategy, covering high school, undergraduate, and graduate-level knowledge.
• Context Window: Supports a 32k context window for processing lengthy scientific texts and complex problems.
• Heterogeneous MoE Architecture: The broader S1-Base series utilizes a Mixture-of-Experts (MoE) architecture to route queries to specialized language models or domain-specific models (e.g., wave, spectrum, protein models).

Good For

• Applications requiring deep scientific understanding and reasoning.
• Tasks involving complex problem-solving in foundational scientific disciplines.
• Processing and analyzing extensive scientific literature.