PeopleofSuperintelligence

Researcher

Niklas Muennighoff

Niklas Muennighoff's research covers open language models, data-limited training and evaluation benchmarks. His MTEB benchmark, published at EACL in May 2023, describes 8 embedding tasks, 58 datasets and 112 languages. At NeurIPS 2023, his paper Scaling Data-Constrained Language Models received an Outstanding Paper Runner-up award. He led 23 researchers at the Allen Institute for AI in building OLMoE, an open language model, and he co-authored Molmo and PixMo, open-weight vision-language models posted to arXiv in September 2024. As of October 2026, he is building Composer, an AI coding model, at Cursor, while he continues his PhD at Stanford.

Now at
Cursor
Works
103
Citations
4,251
h-index
24

Publication numbers from OpenAlex, as of 15 October 2026.

Timeline

  1. 2020

    Second place in Hateful Memes

    At Peking University, he builds an AI system to reduce hate speech. In 2020, his vision-language models finish second among more than 3,300 teams in Meta's Hateful Memes Competition.

  2. 2023

    A benchmark for text embeddings

    MTEB, a text-embedding benchmark he co-writes, covers 8 embedding tasks, 58 datasets and 112 languages. EACL publishes the paper in May 2023.

  3. 2023

    A NeurIPS runner-up paper

    At NeurIPS 2023, his paper Scaling Data-Constrained Language Models receives an Outstanding Paper Runner-up award.

  4. 2024

    OLMoE at the Allen Institute

    He leads 23 researchers at the Allen Institute for AI in building OLMoE, an open language model submitted to arXiv in September 2024.

  5. September 2024

    Co-author of Molmo and PixMo

    He co-authors Molmo and PixMo, open-weight vision-language models posted to arXiv in September 2024.

  6. 2024–2025

    Knight-Hennessy Scholar at Stanford

    In autumn 2024, Stanford admits him to its computer-science PhD program. The 2025 cohort of Knight-Hennessy Scholars includes him.

  7. 2025

    Best paper award for s1

    He co-writes s1: Simple test-time scaling, submitted to arXiv in January 2025. The paper wins a best paper award at the EMNLP and ICLR 2025 Reasoning Workshop.

  8. As of October 2026

    Building Composer at Cursor

    He is building Composer, an AI coding model, at Cursor, while he continues his PhD at Stanford. The Composer 2 technical report, submitted to arXiv in March 2026, lists him among its Cursor Research authors.

Sources and credits

Photo: We have no published photo of Niklas Muennighoff yet. Send us a free-licensed photo.

  1. Niklas Muennighoff muennighoff.github.io. Checked 15 October 2026.
  2. Niklas Muennighoff | Knight-Hennessy Scholars at Stanford University knight-hennessy.stanford.edu. Checked 15 October 2026.
  3. [2210.07316] MTEB: Massive Text Embedding Benchmark arxiv.org. Checked 15 October 2026.
  4. MTEB: Massive Text Embedding Benchmark - ACL Anthology aclanthology.org. Checked 15 October 2026.
  5. NeurIPS 2023 Awards neurips.cc. Checked 15 October 2026.
  6. [2409.02060] OLMoE: Open Mixture-of-Experts Language Models arxiv.org. Checked 15 October 2026.
  7. [2409.17146] Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models arxiv.org. Checked 15 October 2026.
  8. Niklas Muennighoff's Profile | Stanford Profiles profiles.stanford.edu. Checked 15 October 2026.
  9. [2501.19393] s1: Simple test-time scaling arxiv.org. Checked 15 October 2026.
  10. [2603.24477] Composer 2 Technical Report arxiv.org. Checked 15 October 2026.