← All notes

Summary

s1: Simple test-time scaling (2025), by Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, and coauthors, is at https://arxiv.org/abs/2501.19393. The authors curate s1K, a set of 1,000 questions paired with reasoning traces, chosen using three criteria they test in ablations: difficulty, diversity, and quality. They develop budget forcing to control test-time compute by forcefully terminating the model's thinking process or lengthening it by appending Wait when the model tries to end. After supervised finetuning of Qwen2.5-32B-Instruct on s1K and using budget forcing, they report that s1-32B exceeds o1-preview on competition math questions by up to 27% (MATH and AIME24), and that scaling with budget forcing can move AIME24 from 50% to 57%. They say the model, data, and code are open-source at https://github.com/simplescaling/s1. Title, authors, abstract, and comments were taken from the arXiv abs page. The version is v3 (1 Mar 2025), submitted 31 Jan 2025.

Claims

Pointer notes do not quote the paper. See Verification and Sources for the link and license.

Verification

• Format: pointer note only. The paper is not reproduced here.
• Why it matters: It matters because it offers a small, openly described recipe for test-time scaling at a moment when many people were trying to reproduce the behavior of closed reasoning models.
• Reproduced: not reproduced (2026-11-11).
• Related work: Related notes already in the library include Self-Consistency and Self-Adapting Language Models.
• Limits: This is a pointer note only. AIME and MATH comparisons are the authors' reported results from the v3 paper (1 Mar 2025).
• License: arXiv default (non-exclusive distribution license) on the abs page (2026-11-11; http://arxiv.org/licenses/nonexclusive-distrib/1.0/). Submitted 31 Jan 2025 (v1); last revised 1 Mar 2025 (v3). Sources: [S1] https://arxiv.org/abs/2501.19393, v3 (1 Mar 2025). checked_on: 2026-11-11. Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it. The note itself is CC BY 4.0.

When to apply

Use when looking for a simple, openly described test-time scaling recipe and the s1K data release.

Facts

• id: F1 claim: 'The s1 paper curates s1K, a dataset of 1,000 questions paired with reasoning traces, selected for difficulty, diversity, and quality.' confidence: medium date: 2026-11-11 sources: [S1] license: CC-BY-4.0
• id: F2 claim: 'The authors describe budget forcing, which can stop or extend a model''s thinking process, including by appending Wait when the model tries to end.' confidence: medium date: 2026-11-11 sources: [S1] license: CC-BY-4.0
• id: F3 claim: 'After supervised finetuning of Qwen2.5-32B-Instruct on s1K with budget forcing, they report that s1-32B exceeds o1-preview on competition math questions by up to 27% on MATH and AIME24, and that budget forcing can move AIME24 from 50% to 57%.' confidence: medium date: 2026-11-11 sources: [S1] license: CC-BY-4.0

Sources

• [S1] arXiv 2501.19393, https://arxiv.org/abs/2501.19393. Paper license: arXiv default (non-exclusive distribution license). v3 (1 Mar 2025). Accessed 2026-11-11. This note links to the paper and does not reproduce it.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

@misc{iil_note_a3fqd2cd,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{s1: Simple test-time scaling}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2501-19393-s1/},
  note = {Permanent ID iil:note:a3fqd2cd. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2501-19393-s1.md.