---
title: "Niklas Muennighoff — author profile and papers"
description: "Niklas Muennighoff, Stanford University: 9 papers on Elamur Research (2022–2025), h-index 24 — affiliations, topics, co-authors and who cites them in our…"
canonical: "https://research.elamur.ai/free/en/authors/niklas-muennighoff/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/authors/niklas-muennighoff/)

[home](https://md.research.elamur.ai/free/en/) / [authors](https://md.research.elamur.ai/free/en/authors/) / Niklas Muennighoff

# Niklas Muennighoff

Niklas Muennighoff works on large language models and the resources used to train and evaluate them. A recurring thread in the papers here is text embeddings: how to measure them across many tasks and languages, and whether generative decoder models can also serve retrieval. Other work covers multilingual instruction finetuning, openly documented code models, vision-language systems and reasoning at inference time. The methods lean on new benchmarks, released datasets and openly shared model families.

The most cited of these papers is [MTEB: Massive Text Embedding Benchmark](https://md.research.elamur.ai/free/en/papers/muennighoff-2022-mteb-massive-text-embedding-benchmark/) (2022). It gathers 58 datasets in 112 languages into eight task types, each with a fixed evaluation protocol. [C-Pack](https://md.research.elamur.ai/free/en/papers/xiao-2023-c-pack-packed-resources/) (2023) carries that approach to Chinese, pairing a Chinese benchmark with a mined training corpus and the BGE family of encoders. [Crosslingual Generalization through Multitask Finetuning](https://md.research.elamur.ai/free/en/papers/muennighoff-2022-crosslingual-generalization-through-multitask-finetuning/) (2022) finetunes BLOOM and mT5 on prompted task mixtures, including a new 46-language collection, xP3, and yields BLOOMZ and mT0. SGPT (2022) adapts GPT-style decoders to semantic search, and SantaCoder (2023) concerns code models whose data and design choices are disclosed.

The newer papers move toward reasoning and multimodal models. [s1: Simple test-time scaling](https://md.research.elamur.ai/free/en/papers/muennighoff-2025-s1-simple-test-time-scaling/) (2025) asks how little is needed for a model to improve by thinking longer. It uses supervised finetuning of Qwen2.5-32B-Instruct on just 1,000 curated questions with distilled reasoning traces, in place of large-scale reinforcement learning. [Generative Representational Instruction Tuning](https://md.research.elamur.ai/free/en/papers/muennighoff-2024-generative-representational-instruction-tuning/) (2024) trains a single model to handle both generation and embedding, with instructions selecting the mode. Molmo and PixMo (2024) builds open vision-language models from human-collected data instead of outputs of proprietary systems.

Muennighoff is at Stanford University, with papers since 2024. Earlier bylines include Peking University, New York University, Hugging Face, Cohere and the Allen Institute for Artificial Intelligence. Co-authors include [Douwe Kiela](https://md.research.elamur.ai/free/en/authors/douwe-kiela/) and [Hailey Schoelkopf](https://md.research.elamur.ai/free/en/authors/hailey-schoelkopf/), and the work falls mainly under [language models and pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) and [evaluation and benchmarks](https://md.research.elamur.ai/free/en/topics/evaluation-benchmarks/).

9 papers on this site · 2022–2025

current affiliation [Stanford University](https://md.research.elamur.ai/free/en/universities/stanford-university/), Stanford, United States

Niklas Muennighoff has 9 papers on this site, in 2022–2025, most of them on [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/), [Multilingual & cross-lingual NLP](https://md.research.elamur.ai/free/en/topics/multilingual-nlp/), [Evaluation & benchmarks](https://md.research.elamur.ai/free/en/topics/evaluation-benchmarks/); often with [Adam Roberts](https://md.research.elamur.ai/free/en/authors/adam-roberts/), [Alex Gu](https://md.research.elamur.ai/free/en/authors/alex-gu/), [Amanpreet Singh](https://md.research.elamur.ai/free/en/authors/amanpreet-singh/).

[OpenAlex](https://openalex.org/A5000043237) · [ORCID](https://orcid.org/0009-0001-7157-770X) · [**h-index 24**](https://openalex.org/authors/A5000043237) · [103 works](https://openalex.org/works?filter=author.id:A5000043237) · [4 251 citations](https://openalex.org/works?filter=author.id:A5000043237&sort=cited_by_count:desc) OpenAlex, 2026-10-14

[9 works on this site](https://md.research.elamur.ai/free/en/authors/niklas-muennighoff/#papers) · [5 652 citations in our library](https://md.research.elamur.ai/free/en/authors/niklas-muennighoff/cited-by/) · 12 more works in our library

The profile is the library's match of the name; two people with one name are not always told apart.

Elsewhere: [Google Scholar](https://scholar.google.com/citations?view_op=search_authors&mauthors=Niklas+Muennighoff) · [Semantic Scholar](https://www.semanticscholar.org/search?q=Niklas+Muennighoff) · [dblp](https://dblp.org/search?q=Niklas+Muennighoff) · [ACL Anthology](https://aclanthology.org/search/?q=Niklas+Muennighoff)

## Affiliations

- [Stanford University](https://md.research.elamur.ai/free/en/universities/stanford-university/) · Stanford, United States —current · 2024–2026 on 14 papers
- [Allen Institute for Artificial Intelligence](https://md.research.elamur.ai/free/en/labs/allen-institute-for-artificial-intelligence/) · Seattle, United States —2025 on 3 papers
- [National Library of Finland](https://ror.org/03qr6jw57) · Helsinki —2023–2025 on 3 papers
- [Cohere (Canada)](https://md.research.elamur.ai/free/en/companies/cohere/) · San Francisco —2024 on 4 papers
- [Peking University](https://md.research.elamur.ai/free/en/universities/peking-university/) · Beijing, China —2021–2024 on 4 papers
- [Carnegie Mellon University](https://md.research.elamur.ai/free/en/universities/carnegie-mellon-university/) · Pittsburgh, United States —2024 on 3 papers
- [New York University](https://md.research.elamur.ai/free/en/universities/new-york-university/) · New York, United States —2022–2024 on 3 papers
- [University of Washington](https://md.research.elamur.ai/free/en/universities/university-of-washington/) · Seattle, United States —2024 on 2 papers
- [Hugging Face](https://md.research.elamur.ai/free/en/companies/hugging-face/) · New York, United States —2023 on 2 papers
- [University of Cambridge](https://md.research.elamur.ai/free/en/universities/university-of-cambridge/) · Cambridge, United Kingdom —2022–2023 on 2 papers

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 14 more institutions on single bylines are not listed.

## Topics

[Language models & pretraining 3](https://md.research.elamur.ai/free/en/topics/language-models/) · [Multilingual & cross-lingual NLP 1](https://md.research.elamur.ai/free/en/topics/multilingual-nlp/) · [Evaluation & benchmarks 1](https://md.research.elamur.ai/free/en/topics/evaluation-benchmarks/) · [Vision & language 1](https://md.research.elamur.ai/free/en/topics/multimodal/)

## Works by year

### 2020s

[**2025** 1 paper](https://md.research.elamur.ai/free/en/papers/muennighoff-2025-s1-simple-test-time-scaling/) · [**2024** 3 papers](https://md.research.elamur.ai/free/en/authors/niklas-muennighoff/year/2024/) · [**2023** 2 papers](https://md.research.elamur.ai/free/en/authors/niklas-muennighoff/year/2023/) · [**2022** 3 papers](https://md.research.elamur.ai/free/en/authors/niklas-muennighoff/year/2022/)

## Papers

The most cited first (OpenAlex and our library).

1. **[MTEB: Massive Text Embedding Benchmark](https://md.research.elamur.ai/free/en/papers/muennighoff-2022-mteb-massive-text-embedding-benchmark/)** *(2022 · 1 021 citations)* Text embedding models are usually tested on only a few datasets, most often semantic textual similarity (STS).
2. **[C-Pack: Packed Resources For General Chinese Embeddings](https://md.research.elamur.ai/free/en/papers/xiao-2023-c-pack-packed-resources/)** *(2023 · 920 citations)* Chinese has lagged behind English in general-purpose text embeddings because it lacks large public training sets, competitive models and a broad evaluation suite.
3. **[Crosslingual Generalization through Multitask Finetuning](https://md.research.elamur.ai/free/en/papers/muennighoff-2022-crosslingual-generalization-through-multitask-finetuning/)** *(2022 · 692 citations)* Multitask prompted finetuning (MTF) helps large language models handle unseen tasks zero-shot, but earlier studies stayed almost entirely within English.
4. **[s1: Simple test-time scaling](https://md.research.elamur.ai/free/en/papers/muennighoff-2025-s1-simple-test-time-scaling/)** *(2025 · 584 citations)* The paper looks for the least machinery needed to reproduce what OpenAI's o1 showed: a model that gets more accurate when it is allowed to think longer at inference time.
5. **[SGPT: GPT Sentence Embeddings for Semantic Search](https://md.research.elamur.ai/free/en/papers/muennighoff-2022-sgpt-gpt-sentence-embeddings-semantic/)** *(2022 · 187 citations)* Large GPT-style decoder models lead many language tasks, but semantic search and sentence embeddings have remained the domain of much smaller BERT-like encoders.
6. **[Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models](https://md.research.elamur.ai/free/en/papers/deitke-2024-molmo-pixmo-open-weights-open/)** *(2024 · 184 citations)* The strongest vision-language models (VLMs) are closed. Most capable open-weight alternatives are trained on captions or answers produced by proprietary systems such as GPT-4V. In effect they distill closed models, so it remains unclear how to build a strong VLM from scratch. The authors release Molmo, a family of…
7. **[Generative Representational Instruction Tuning](https://md.research.elamur.ai/free/en/papers/muennighoff-2024-generative-representational-instruction-tuning/)** *(2024 · 165 citations)* LLMs are usually specialised. A model tuned for text generation produces weak embeddings, and a model tuned for embeddings loses its ability to generate. This paper introduces GRIT (generative representational instruction tuning), which trains one LLM on both kinds of data. The instruction format tells the model…
8. **[SantaCoder: don't reach for the stars!](https://md.research.elamur.ai/free/en/papers/allal-2023-santacoder-don-t-reach-stars/)** *(2023 · 164 citations)* Most code LLMs are built with little transparency: training data, preprocessing, and design choices are rarely disclosed.
9. **[Scaling Laws with Vocabulary: Larger Models Deserve Larger Vocabularies](https://md.research.elamur.ai/free/en/papers/tao-2024-scaling-laws-vocabulary-larger-models/)** *(2024 · 42 citations)* Standard scaling laws for LLMs predict loss from parameter count and training data but treat vocabulary size as fixed.

## Also in our library

12 more works of theirs are in our library without a page here yet; they will appear as we publish them.

## Co-authors here

12 of 45 co-authors with a page, the most papers together first.

- [Adam Roberts](https://md.research.elamur.ai/free/en/authors/adam-roberts/)
- [Alex Gu](https://md.research.elamur.ai/free/en/authors/alex-gu/)
- [Amanpreet Singh](https://md.research.elamur.ai/free/en/authors/amanpreet-singh/)
- [Carolyn Jane Anderson](https://md.research.elamur.ai/free/en/authors/carolyn-jane-anderson/)
- [Chenghao Mou](https://md.research.elamur.ai/free/en/authors/chenghao-mou/)
- [Christopher Akiki](https://md.research.elamur.ai/free/en/authors/christopher-akiki/)
- [Christopher Clark](https://md.research.elamur.ai/free/en/authors/christopher-clark/)
- [Defu Lian](https://md.research.elamur.ai/free/en/authors/defu-lian/)
- [Douwe Kiela](https://md.research.elamur.ai/free/en/authors/douwe-kiela/)
- [Dragomir Radev](https://md.research.elamur.ai/free/en/authors/dragomir-radev/)
- [Furu Wei](https://md.research.elamur.ai/free/en/authors/furu-wei/)
- [Hailey Schoelkopf](https://md.research.elamur.ai/free/en/authors/hailey-schoelkopf/)

## Cited by

Papers of our library cite their works 5 652 times (of the 140 884 papers whose bibliographies we read); [299 citing papers with a page here →](https://md.research.elamur.ai/free/en/authors/niklas-muennighoff/cited-by/).