PeopleofSuperintelligence

Researcher

Anton Bakhtin

Anton Bakhtin is an AI researcher whose work spans game-playing agents, language models and physical reasoning. Among the work they co-authored is CICERO, a 2022 paper on human-level play in the game of Diplomacy. In 2020 they co-wrote ReBeL with Noam Brown, a framework that pairs self-play reinforcement learning with search for imperfect-information games. ICLR 2023 gave their no-press Diplomacy paper an Honorable Mention. Their most recent arXiv preprints, from September and October 2026, cover agentic reasoning and self-improving agents.

Works
27
Citations
2,475
h-index
11

Publication numbers from OpenAlex, as of 15 October 2026.

Timeline

  1. 2016

    Quantizing acoustic models

    The paper on quantizing neural networks to 8-bit integers, posted to arXiv in 2016, lists Anton Bakhtin among its authors.

  2. 2019

    A physical reasoning benchmark

    PHYRE, a benchmark of 2D mechanics puzzles for physical reasoning, is posted to arXiv in 2019 with Anton Bakhtin among its authors.

  3. 2020

    ReBeL for imperfect-information games

    ReBeL, a framework that pairs self-play reinforcement learning with search for imperfect-information games, is posted to arXiv in July 2020 with Noam Brown and Anton Bakhtin among its authors.

  4. 2022

    CICERO and human-level Diplomacy

    CICERO, a 2022 paper on human-level play in the game of Diplomacy, lists Anton Bakhtin among its authors.

  5. 2022–2023

    An ICLR honorable mention

    A no-press Diplomacy paper, posted to arXiv in October 2022 with Anton Bakhtin among its authors, receives an ICLR 2023 Honorable Mention.

  6. 2023

    Measuring representation of global opinions

    A paper on whose opinions language models resemble, posted to arXiv in June 2023, lists Anton Bakhtin among its authors.

  7. September 2026

    Agentic meta-reasoning for long runs

    Thinking Before Thinking, which introduces agentic meta-reasoning for long AI agent runs, is posted to arXiv in September 2026 with Anton Bakhtin among its authors.

  8. October 2026

    Self-improvement through experience

    Agent Plasticity, posted to arXiv in October 2026 with Anton Bakhtin among its authors, defines agent plasticity as the efficiency with which an agent turns experience into gains.

Sources and credits

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

  1. [1607.04683] On the efficient representation and execution of deep acoustic models arxiv.org. Checked 15 October 2026.
  2. [1908.05656] PHYRE: A New Benchmark for Physical Reasoning arxiv.org. Checked 15 October 2026.
  3. [2007.13544] Combining Deep Reinforcement Learning and Search for Imperfect-Information Games arxiv.org. Checked 15 October 2026.
  4. ‘Diplomacy AI’ directory · Gwern.net gwern.net. Checked 15 October 2026.
  5. [2210.05492] Mastering the Game of No-Press Diplomacy via Human-Regularized Reinforcement Learning and Planning arxiv.org. Checked 15 October 2026.
  6. ICLR 2023 Awards iclr.cc. Checked 15 October 2026.
  7. [2306.16388] Towards Measuring the Representation of Subjective Global Opinions in Language Models arxiv.org. Checked 15 October 2026.
  8. [2609.38147] Thinking Before Thinking: Scaling Agentic Inference Through Meta-Reasoning arxiv.org. Checked 15 October 2026.
  9. [2610.08902] Agent Plasticity: Measuring Self-Improvement Through Experience arxiv.org. Checked 15 October 2026.