ABAB Pedia

Koray Kavukcuoglu

Koray Kavukcuoglu is an artificial intelligence researcher and engineer who joined DeepMindGoogle DeepMindGoogle DeepMind is Google’s artificial intelligence research organization, formed in April 2023 by merging DeepMind, which Google acquired in 2014, with the Brain team from Google Research, and led by CEO Demis Hassabis. It developed AlphaGo, AlphaFold, and the Gemini family of models, and Hassabis and John Jumper received the 2024 Nobel Prize in Chemistry for AlphaFold’s protein structure prediction work.Open the full entry in 2012 and worked on research such as DQN and WaveNet; he has been GoogleGoogleGoogle is an American technology company founded by Larry Page and Sergey Brin and incorporated in 1998. It became an Alphabet subsidiary in 2015. Its businesses include internet search, advertising, software platforms, cloud computing, and artificial intelligence.Open the full entry’s chief AI architect since June 2025 and, since August 2026, senior vice president of Google DeepMind, overseeing Gemini model development and the lab’s day-to-day operations.

Contents21 sections
Key facts

Undergraduate Studies at Middle East Technical University

He was born in Türkiye.⁠[1][2]

He studied as an undergraduate at Middle East Technical University (ODTÜ in Turkish), graduating from its aerospace engineering department in 1999.⁠[3][4][1]

Engineer at Roketsan

He worked as an engineer at the Turkish company Roketsan; in 2003 he completed a master’s degree in the same department at Middle East Technical University.⁠[4][3]

He began as an aerospace engineer before turning to computer science.⁠[5]

Doctoral Thesis at New York University

At New York UniversityNew York UniversityNew York University is a private university in New York City, founded in 1831, and one of the largest private universities in the United States.Open the full entry he earned a master’s degree in computer science and studied for a doctorate under Yann LeCun; he also interned at Siemens, NEC Laboratories America and Google.⁠[4][5][3]

His doctoral thesis, Learning Feature Hierarchies for Object Recognition, was submitted to the Department of Computer Science at New York University, with Yann LeCun as his adviser; another account gives 2010 as the year he completed his doctorate.⁠[6][1]

In the thesis acknowledgements he thanks his adviser, Yann LeCun, for his continuous guidance and calls it a great privilege to have worked with and learned from him, thanks Ronan Collobert and Jason Weston for the opportunity to work with them, and says that discussions and collaborations in LeCun’s laboratory contributed significantly to the work.⁠[6]

Researcher at NEC Laboratories America

He worked as a researcher at NEC Laboratories America.⁠[4][3]

Joining DeepMind

He joined DeepMind as a researcher and started its deep learning team there; he subsequently served as Research Scientist, Director of Research and VP of Research.⁠[4][5][3][8]

DQN Paper in Nature

The deep Q-network (DQN) presented in the paper learned to play a range of Atari 2600 games from raw screen pixels alone; it outperformed previous machine learning methods in 43 of the 49 games tested and reached more than 75% of the performance of professional human testers in more than half of the games.⁠[10]

WaveNet Paper

The paper WaveNet: A Generative Model for Raw Audio was submitted to arXiv, with him as the last of its nine authors; WaveNet is a probabilistic, autoregressive neural network that generates raw audio, reaching near-human naturalness in text-to-speech and also producing music.⁠[11]

Sundar Pichai credits him with leading the way on breakthroughs such as WaveNet and DQN, and his Google author profile says that his team pioneered major research breakthroughs such as DQN and WaveNet.⁠[12][8]

Election as a Fellow of the Royal Academy of Engineering

The annual general meeting of the Royal Academy of Engineering elected 72 new Fellows, among them him, as VP of Research & Technology at DeepMind; the new Fellows were formally admitted at a ceremony in London on November 8 of that year, where each signed the roll book.⁠[13]

The Academy’s citation recognises his contributions to artificial intelligence, particularly deep learning, unsupervised learning, generative models, deep reinforcement learning and the application of AI to real-world problems; it finds that he combines deep technical knowledge, research leadership and the management skills to run a large research organisation, and credits him with “more than 100,000 citations and a stellar international reputation.”⁠[14]

Leading the Scientific Board of Google DeepMind

When Demis Hassabis announced the formation of Google DeepMind, he said that a new Scientific Board would oversee the unit’s research progress and direction, led by Kavukcuoglu and with representatives from across the organisations; Zoubin Ghahramani would join the research leadership team reporting to Kavukcuoglu.⁠[15]

Chief Technology Officer of DeepMind

He became chief technology officer of DeepMind, the company’s top technical leader.⁠[4][3]

As of June 2025, he headed the teams working on AI systems such as Bard, Gemini and AlphaFold and played a central role in setting the company’s AI strategy and technological direction.⁠[3]

Announcing Gemini 2.5

As CTO of Google DeepMind he announced Gemini 2.5Gemini (AI model)Gemini is a family of multimodal AI models developed by Google DeepMind and other Google teams, first announced in December 2023. Its variants cover on-device use, general reasoning and agentic tasks, with access through applications and developer platforms.Open the full entry, describing it as a thinking model that reasons before responding; its first release, Gemini 2.5 Pro Experimental, debuted at number one on LMArena by a significant margin, shipped with a 1 million token context window (2 million planned) and scored 63.8% on SWE-Bench Verified with a custom agent setup; it was first available in Google AI Studio and, for Gemini Advanced users, in the Gemini app, with Vertex AI to follow.⁠[16]

Google’s Chief AI Architect

In a memo to employees, Sundar Pichai announced that he would become Google’s chief AI architect, a new senior vice president position reporting directly to Pichai, while remaining chief technology officer of Google DeepMind.⁠[5][17]

Pichai gave the purpose of the change as to “accelerate how we bring our world-leading models into our products,” with the aim of “more seamless integration, faster iteration, and greater efficiency”; Kavukcuoglu had already been working with Google product leaders to put its models into new products, and the new role formalised that work.⁠[17][5]

Co-Announcing Gemini 3

With Sundar Pichai and Demis Hassabis he co-signed the announcement of Gemini 3: Gemini 3 Pro was released in preview that day, and Gemini 3 began rolling out to everyone in the Gemini app, to Google AI Pro and Ultra subscribers in AI Mode in Search, to developers in the Gemini API in AI Studio, Google Antigravity and Gemini CLI, and to enterprises in Vertex AI and Gemini Enterprise; the enhanced reasoning mode Deep Think went first to safety testers, with Google AI Ultra subscribers to follow; and Google Antigravity was introduced as a new agentic development platform.⁠[18]

Senior Vice President of Google DeepMind

He took over the day-to-day operations of Google DeepMind without a stand-alone chief executive title and kept his role as Google’s chief AI architect.⁠[19][20]

Hassabis recalled that he and Kavukcuoglu had worked together for over 13 years, since the early days of DeepMind, and called him “one of the world's foremost AI experts”; Pichai added: “I look forward to seeing him lead GDM into this next chapter.”⁠[12]

After the reshuffle he was the only one of Gemini’s co-leads still at Google; Yann LeCun describes him as a great engineer who built rare research-management skills on the way up.⁠[2]

Remarks on Gemini 4 at The Information’s AI Agenda Live

At The Information’s AI Agenda Live conference he said that Google wanted to release Gemini 4 “much earlier” than the end of the year, that the model was in post-training and being tried in Antigravity, Google’s coding tool; he gave no release date. It was his first media appearance since taking over Google DeepMind.⁠[21]

Asked whether Google had fallen behind, he said he had “the utmost trust in the team” and that “it's a certainty that we are always gonna be at the frontier.”⁠[21]

Announcing Gemini 4 Argon

As senior vice president of Google DeepMind and Google’s chief AI architect, he announced the frontier model Gemini 4 Argon, aimed at software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defence; it can sustain long multi-step tasks, and its output token limit rose to 1 million from 64,000.⁠[22]

The Fairwind Program, announced at the same time, gave Argon first to a group of trusted cyber defenders, with availability for developers, enterprises and consumers starting with paid API customers and Google AI Ultra subscribers; the introductory price was $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper, rising to $4 and $20 after the introductory period.⁠[22]

Announcing General Availability of Gemini 4 Carbon

He announced that Gemini 4 Carbon was being made available to developers, enterprises, scientists and consumers, calling it Google’s strongest model, with frontier performance in cybersecurity and long-horizon agentic workflows and strengths in complex knowledge work, software engineering and scientific reasoning. Carbon had been released two weeks earlier to cybersecurity defenders through the Fairwind Program: Google announced Carbon on October 28, 2026, saying it would go first to cyber defenders and then to developers, businesses and consumers.⁠[23][24][25]

The post advises that Argon remains the better choice for most software engineering and knowledge work, while Carbon targets harder tasks such as complex multi-step coding and open-ended research and uses more tokens to explore thoroughly and run extended verification loops; Google says Carbon ranks first on CursorBench 4.0, PerplexityPerplexityPerplexity is a San Francisco–based artificial intelligence company founded in 2022, known for its “answer engine” that gives answers with cited sources, and it has also launched products such as the Comet browser.Open the full entry’s WANDR and Zapier’s AutomationBench at more than two times lower cost per task than the next-best external model, and second on the Multimodal Knowledge Agent benchmark at more than six times lower average cost per task than the top model.⁠[23]

Paying developers and enterprises can use Carbon through the API in Google AI Studio and Gemini for business at the same introductory price as Argon, $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after November 30, 2026, and the paid tier of the Gemini API can also be used in AI Studio and Android Studio; it is rolling out to all Google AI Ultra subscribers in Google Antigravity, the Gemini app and AI Mode in Search; scientists can apply for six months of free access to Argon and Carbon through Antigravity’s highest usage tier; and cybersecurity defenders can apply to the Fairwind Program for Carbon for cyber.⁠[23][25]

The post gives examples: in the Scan for Good programme of Wiz, Inc.Wiz (cloud security company)Wiz is a cloud and AI security platform company headquartered in New York, founded in January 2020 by Assaf Rappaport, Ami Luttwak, Yinon Costica and Roy Reznik. Google announced an all-cash agreement to acquire Wiz for $32 billion in March 2025 and completed the deal on March 11, 2026; Wiz joined Google Cloud and kept its brand.Open the full entry, Carbon found zero-day vulnerabilities in two major SaaS providers in under two hours, which Google says compressed weeks of vulnerability research into a single afternoon; in science, agents of the AlphaProof Code framework built by researchers used Carbon alongside other models to generate Lean proof code, helping prove key security properties of the code implementing SignalSignalSignal is an end-to-end encrypted messaging app run by a nonprofit organization; built on the open-source Signal Protocol, it offers text messaging, voice calls and other features, and its client and server code is publicly released.Open the full entry’s cryptographic protocol in weeks, a task that historically takes expert teams years; other examples include enzyme assay analysis and protein homolog discovery, where it increased retrieved homologs about 40-fold on average across seven low-homolog families.⁠[23][25]

Google also previewed /swarm in Antigravity, which coordinates a team of agents, with Carbon taking deep-reasoning roles such as architectural design, coordination and auditing; on safety, Google says Carbon blocked 99% of attack trajectories from an adaptive red-teaming agent, lowering the per-attempt breach rate to 0.8% from 3.1% on Argon, and recorded a 0% prompt injection success rate on the Gray Swan IPI benchmark, while acknowledging that prompt injection robustness can still improve.⁠[23]

Family and Personal Life

He is married and has daughters; his doctoral thesis is dedicated “To my family,” and in its acknowledgements he says that, most importantly, he is indebted to his wife, his daughters and all his family, without whose endless support he could not have completed the doctorate.⁠[6]

Academic Standing

As of June 2025, his work had received nearly 270,000 academic citations over the preceding eight years.⁠[3]

Related Organizations, Websites, and Public Accounts

Google DeepMind: Official website: https://deepmind.google/ (opens in a new window).⁠[15]

Royal Academy of Engineering: Official website: https://raeng.org.uk/ (opens in a new window).⁠[14]

Sources28 sources

  1. DeepMind to be led by Turk as Google shakes up AI division (opens in a new window)dailysabah.com↑
  2. Google just moved its AI crown from London to California. Meet the man who caught it (opens in a new window)thenextweb.com↑
  3. Google’ı yöneten eski ROKETSAN mühendisi! [The former ROKETSAN engineer who runs Google!] (opens in a new window)shiftdelete.net↑
  4. Google'ın Gemini 3 ile Yapay Zekâda Yeni Bir Çağ Başlatmasını Sağlayan İsimlerden Koray Kavukçuoğlu Kimdir? [Who Is Koray Kavukçuoğlu, One of the Figures Who Enabled Google to Start a New Era in AI with Gemini 3?] (opens in a new window)webtekno.com↑
  5. Google names new chief AI architect to advance developments (opens in a new window)semafor.com↑
  6. Learning Feature Hierarchies for Object Recognition (opens in a new window)cs.nyu.edu↑
  7. Torch7: A Matlab-like Environment for Machine Learning (opens in a new window)infoscience.epfl.ch↑
  8. Koray Kavukcuoglu (opens in a new window)blog.google↑
  9. Human-level control through deep reinforcement learning (opens in a new window)ideas.repec.org↑
  10. From Pixels to Actions: Human-level control through Deep Reinforcement Learning (opens in a new window)research.google↑
  11. WaveNet: A Generative Model for Raw Audio (opens in a new window)arxiv.org↑
  12. The next chapter of our AI momentum (opens in a new window)blog.google↑
  13. Royal Academy of Engineering welcomes 72 new Fellows (opens in a new window)raeng.org.uk↑
  14. Koray Kavukcuoglu FREng (opens in a new window)raeng.org.uk↑
  15. Announcing Google DeepMind (opens in a new window)deepmind.google↑
  16. Gemini 2.5: Our most intelligent AI model (opens in a new window)blog.google↑
  17. Google adds chief AI architect to accelerate development of AI products (opens in a new window)nbcsandiego.com↑
  18. A new era of intelligence with Gemini 3 (opens in a new window)blog.google↑
  19. Demis Hassabis steps down from Google DeepMind CEO role amid a major AI leadership shake-up (opens in a new window)fortune.com↑
  20. Turkish engineer named to lead Google DeepMind (opens in a new window)hurriyetdailynews.com↑
  21. Google wants Gemini 4 out ‘much earlier’ than the end of the year (opens in a new window)thenextweb.com↑
  22. Gemini 4 Argon (opens in a new window)blog.google↑
  23. Gemini 4 Carbon is now available (opens in a new window)blog.google↑
  24. Google announces Gemini 4 Carbon with ‘significant gains across coding and agentic tasks’ (opens in a new window)9to5google.com↑
  25. 谷歌向付费开发者与AI Ultra用户开放Gemini 4 Carbon [Google opens Gemini 4 Carbon to paying developers and AI Ultra users] (opens in a new window) (in Chinese)ababnews.com↑
  26. Our story (opens in a new window)about.google↑
  27. koray kavukcuoglu (@koraykv) / X (opens in a new window)x.com↑
  28. koraykv (koray kavukcuoglu) · GitHub (opens in a new window)github.com↑