Find open-source science resources

Pretrained time series foundation model for long-horizon forecasting across diverse scientific domains including climate variables, biomedical signals, and physical observations; decoder-only Transformer architecture with strong zero-shot generalization (19.8K+ stars, Apache 2.0, 2024-2025)

Foundation model for tabular data that predicts on unseen real-world tables in a single forward pass, achieving accurate small-data classification and regression without task-specific training; widely applicable to scientific datasets with limited samples (7.4K+ stars, 2022-2026)

Open-source scientific multimodal foundation model built on a 235B MoE LLM and 6B vision encoder, continually pretrained on 5T tokens including 2.5T scientific-domain tokens, with strong results across chemistry, materials, life science, and earth science benchmarks (2025)

Scikit-learn compatible tabular foundation model for zero-shot classification and regression on mixed-type tabular datasets via in-context learning; applicable to diverse scientific datasets (1.8K+ stars, Apache 2.0)

Pretrained time series foundation model for zero-shot forecasting across diverse scientific and real-world domains; tokenizes continuous time series into discrete bins to train transformer language models on large-scale corpora, achieving strong zero-shot generalization and competitive performance with task-specific supervised models on climate, energy, and health benchmarks (5.3K+ stars, Apache 2.0, 2024-2026)

Scientific taste learning framework showing AI can judge and propose research ideas with long-term impact; trains Scientific Judge as a generative reward model and Scientific Thinker as an ideation policy using Reinforcement Learning from Community Feedback (RLCF) on large-scale citation signals, with SciJudgeBench and released HuggingFace model weights (425+ stars, Apache 2.0, 2026)

First multi-domain generative foundation model for the natural sciences built on a unified scientific grammar, encoding proteins, antibodies, small molecules, chemical reactions, materials, and their spatial interactions into a shared token vocabulary; enables unified generation, prediction, and design across domains under a purely autoregressive paradigm (134+ stars, Apache 2.0, 2026)

Universal time series forecasting via the UNI2TS library, training a single transformer with shared self-attention and specialized mixture-of-experts feed-forward blocks to achieve strong zero-shot generalization across heterogeneous domains including energy, weather, transportation, and health time series (1.6K+ stars, Apache 2.0, 2024-2026)

Scientific foundation model and AI research copilot for idea generation, cross-disciplinary connection discovery, and hypothesis formation; trained with a decoupled reward-comment RL architecture and achieves GPT-4o-competitive novelty/rationale on STEM and social-science idea-generation benchmarks (270+ stars, MIT License, 2026)

Open language model for mathematics (7B/34B) trained on Proof-Pile-2, outperforming Minerva at equal scale on MATH benchmark, with tool use and formal theorem proving in Lean without finetuning (EleutherAI, ICLR 2024)

Large language model for science

Mathematical reasoning