PeopleofSuperintelligence

Researcher

Piotr Padlewski

Also known as prazek

Piotr Padlewski is a member of technical staff at Anthropic whose background spans multimodal language models, vision transformers and compilers. While a BA student at the University of Warsaw, he presented C++ devirtualization work at the 2016 LLVM Developers' Meeting, and a 2020 paper on the invariance of virtual pointers in LLVM describes the model now in upstream LLVM/Clang. From 2018 to 2023, he was a researcher at Google DeepMind in Zurich, where he worked on Pathways, ViT-22B, PaLI, NaViT and Gemini. At Reka from 2023 to 2025, he was among the authors of the Reka Core, Flash, and Edge report and first author of the Vibe-Eval benchmark in 2024. He joined Anthropic in Zurich in February 2025 and works on multimodal LLM and VLM training.

Now at
Anthropic
Works
12
Citations
585
h-index
8
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2013–2017

    BA at the University of Warsaw

    He studies for a BA at the University of Warsaw from 2013 to 2017. From 2016 to 2017, he also interns at Google, working on LLVM.

  2. 2016–2020

    Devirtualization in LLVM

    He travels to the 2016 LLVM Developers' Meeting in San Jose on an LLVM Foundation travel grant and writes up his C++ devirtualization work for the LLVM Project Blog in 2017. A 2020 paper with Krzysztof Pszeniczny and Richard Gregory Smith presents a model for the invariance of virtual pointers in LLVM; it gives an average 0.8% speedup on real-world C++ programs and is part of upstream LLVM/Clang behind the -fstrict-vtable-pointers flag.

  3. 2018–2023

    Researcher at Google DeepMind

    He works as a researcher at Google DeepMind in Zurich from 2018 to November 2023, on Pathways, ViT-22B, PaLI, NaViT and Gemini. In 2022, he is among the authors of PaLI, a Pathways Language and Image model that handles language and vision jointly.

  4. 2023

    ViT-22B and NaViT

    He is among the authors of ViT-22B in February 2023, a Vision Transformer with 22 billion parameters. In July 2023, he is among the authors of NaViT, which packs sequences during training to process images of arbitrary resolution and aspect ratio.

  5. November 2023–February 2025

    Member of technical staff at Reka

    He is a member of technical staff at Reka from November 2023 to February 2025, working on backend systems, C++, performance optimizations, compilers and multimodal language models. In April 2024, he is among the Reka Team authors of the Reka Core, Flash, and Edge report, which covers multimodal models trained from scratch that process text, images, video and audio.

  6. May 2024

    The Vibe-Eval benchmark

    He is first author of Vibe-Eval, an open benchmark and framework for evaluating multimodal chat models, in May 2024. It has 269 visual understanding prompts, 100 of them hard, with gold-standard responses written by experts.

  7. January 2025

    Humanity's Last Exam

    He is among the authors of Humanity's Last Exam, submitted in January 2025. The benchmark has 2,500 questions across mathematics, humanities and the natural sciences, developed by subject-matter experts.

  8. Since February 2025

    Multimodal training at Anthropic

    He joins Anthropic in Zurich as a member of technical staff in February 2025, working on multimodal LLM and VLM training.

Career

  1. Since 2025AnthropicMember of technical staff
  2. 2023–2025Reka AIMember of technical staff
  3. 2018–2023Google DeepMindResearcher
  4. 2016–2017GoogleIntern

Education

  1. 2017University of WarsawBA

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. PaLI: A Jointly-Scaled Multilingual Language-Image Model2022 · 235 citations
  2. Scaling Vision Transformers to 22 Billion Parameters2023 · 112 citations
  3. A benchmark of expert-level academic questions to assess AI capabilities2026 · Nature · 61 citations
  4. On Scaling Up a Multilingual Vision and Language Model2024 · IEEE Computer Society Conference on Computer Vision and Pattern Recognition · 52 citations
  5. Patch n’ Pack: NaViT, a Vision Transformer for any Aspect Ratio and Resolution2023 · neural information processing systems · 45 citations

From OpenAlex, as of 21 October 2026.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Piotr Padlewski. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Piotr Padlewski. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.
  3. 22 of 133 people shaping AI with a former OpenAI role now work at AnthropicIncludes Anthropic. Our estimates of where OpenAI recruits came from and where 133 people with former OpenAI roles work now, including the companies and labs they founded.
  4. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Anthropic. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

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

  1. Piotr Padlewski | alphaXiv alphaxiv.org. Checked 20 October 2026.
  2. Devirtualization in LLVM and Clang - The LLVM Project Blog blog.llvm.org. Checked 20 October 2026.
  3. Modeling the Invariance of Virtual Pointers in LLVM research.google. Checked 20 October 2026.
  4. PaLI: A Jointly-Scaled Multilingual Language-Image Model arxiv.org. Checked 20 October 2026.
  5. Scaling Vision Transformers to 22 Billion Parameters arxiv.org. Checked 20 October 2026.
  6. Patch n' Pack: NaViT, a Vision Transformer for any Aspect Ratio and Resolution arxiv.org. Checked 20 October 2026.
  7. Reka Core, Flash, and Edge: A Series of Powerful Multimodal Language Models arxiv.org. Checked 20 October 2026.
  8. Vibe-Eval: A hard evaluation suite for measuring progress of multimodal language models arxiv.org. Checked 20 October 2026.
  9. Piotr Padlewski (Reka AI) — ThursdAI guest thursdai.news. Checked 20 October 2026.
  10. Humanity's Last Exam arxiv.org. Checked 20 October 2026.
  11. Piotr Padlewski - Google Scholar scholar.google.com. Checked 20 October 2026.