Researcher

Cyril Zhang

Cyril Zhang is a computer scientist whose work connects learning theory, transformers, and language models. He grew up in Toronto, earned a computer-science bachelor's degree at Yale, and completed a Princeton Ph.D. under Elad Hazan. His research ranges from transformers that emulate automata to Microsoft's Phi-3 and Phi-4 models and language-model self-improvement. He also coauthored OpenAI's GPT-5 system card. As of October 2026, he is a researcher at OpenAI after earlier work at Microsoft Research NYC.

Now at
OpenAI
Works
46
Citations
610
h-index
13
Badges

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

  1. 2013

    A fourth-place Yale team

    He competes for Yale Team 1 at the ACM International Programming Contest in 2013, and the team finishes fourth overall.

  2. 2015–2020

    Computer science at Yale and Princeton

    He grows up in Toronto and earns a computer-science B.S. at Yale in 2015. At Princeton, he completes a computer-science Ph.D. in 2020 with Elad Hazan as his adviser.

  3. 2023

    Transformers learn shortcuts to automata

    In 2023, he coauthors “Transformers Learn Shortcuts to Automata,” an ICLR paper on how low-depth transformers represent finite-state automata computations.

  4. 2024

    Phi-3 for phones

    In 2024, he coauthors Microsoft Research's Phi-3 technical report, which introduces Phi-3-mini, a 3.8-billion-parameter language model small enough to run on a phone.

  5. December 2024

    A 14-billion-parameter Phi-4

    In December 2024, he coauthors the Phi-4 technical report, which presents a 14-billion-parameter language model trained with a focus on data quality.

  6. 2025

    Self-improvement through sharpening

    In 2025, he coauthors the ICLR paper “Self-Improvement in Language Models: The Sharpening Mechanism.” The work studies using a model as its own verifier during post-training.

  7. August 2025

    The GPT-5 system card

    In August 2025, he coauthors the OpenAI GPT-5 system card, released alongside the model launch. It describes a unified system with a fast model, a deeper reasoning model, and a real-time router.

  8. As of October 2026

    Researcher at OpenAI

    As of October 2026, he is a researcher at OpenAI after earlier work at Microsoft Research NYC. His research focuses on deep-learning theory at frontier scales.

Career

  1. NowOpenAIResearcher
  2. 2020–2025Microsoft

Education

  1. 2020Princeton UniversityPhD
  2. 2015Yale UniversityB.S.

Most-cited works

  1. Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone2024 · 164 citationsWith Adil Salim, Ahmed Awadallah, Caio César Teodoro Mendes, Ce Liu, Corby Rosset and Ronen Eldan
  2. Spectral Filtering for General Linear Dynamical Systems2018 · 55 citationsWith Elad Hazan
  3. Learning Linear Dynamical Systems via Spectral Filtering2017 · 54 citationsWith Elad Hazan
  4. Efficient Regret Minimization in Non-Convex Games2017 · 44 citationsWith Elad Hazan
  5. Phi-4 Technical Report2024 · 31 citationsWith Adil Salim, Caio César Teodoro Mendes, Dingli Yu, Ronen Eldan, Shital Shah and Sébastien Bubeck

Sources and credits

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

  1. Yale hosts and competes in major programming competition news.yale.edu. Checked 19 October 2026.
  2. Cyril Zhang cyrilzhang.com. Checked 19 October 2026.
  3. Transformers Learn Shortcuts to Automata arxiv.org. Checked 19 October 2026.
  4. Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone microsoft.com. Checked 19 October 2026.
  5. Phi-4 Technical Report microsoft.com. Checked 19 October 2026.
  6. Self-Improvement in Language Models: The Sharpening Mechanism arxiv.org. Checked 19 October 2026.
  7. OpenAI GPT-5 System Card arxiv.org. Checked 19 October 2026.