The year/Independent research

Paper 2509.21320

SciReasoner: Laying the Scientific Reasoning Ground Across Disciplines

Published
Sep 2025
Research lab
Independent
Citations
9
GitHub
90 stars

01 In brief

Summary

SciReasoner is a scientific reasoning foundation model that aligns natural language with heterogeneous scientific representations.

It is pretrained on a 206B-token corpus (scientific text, pure sequences, sequence-text pairs) and post-trained via SFT on 40M instructions, annealed cold-start bootstrapping for chain-of-thought, and reinforcement learning with task-specific reward shaping.

The model supports five capability families covering 103 tasks: scientific translation, text/knowledge extraction, property prediction, property classification, and sequence generation/design.

Key innovations include adaptive reasoning (distinguishing 'instant' vs 'thinking' tasks), task-grouped rewarding (distance-based, matching-based, tool-verified), and reward softening (continuous [0,1] signals).

Compared to specialist systems, it broadens instruction coverage, improves cross-domain generalization, and enhances fidelity.

The model achieves state-of-the-art performance on 54 tasks and top-2 on 101 tasks.

Ablation shows pretraining yields consistent gains (median +15.0 pp, mean +22.0 pp on nine tasks; ESOL RMSE drops 98.7%).

The model, datasets, and code are open-sourced.

02 From the paper

Abstract

We present a scientific reasoning foundation model that aligns natural language with heterogeneous scientific representations. The model is pretrained on a 206B-token corpus spanning scientific text, pure sequences, and sequence-text pairs, then aligned via SFT on 40M instructions, annealed cold-start bootstrapping to elicit long-form chain-of-thought, and reinforcement learning with task-specific reward shaping, which instills deliberate scientific reasoning. It supports four capability families, covering up to 103 tasks across workflows: (i) faithful translation between text and scientific formats, (ii) text/knowledge extraction, (iii) property prediction, (iv) property classification, (v) unconditional and conditional sequence generation and design. Compared with specialist systems, our approach broadens instruction coverage, improves cross-domain generalization, and enhances fidelity. We detail data curation and training and show that cross-discipline learning strengthens transfer and downstream reliability. The model, instruct tuning datasets and the evaluation code are open-sourced at https://huggingface.co/SciReason and https://github.com/open-sciencelab/SciReason.