--- title: Koray Kavukcuoglu on frontier models, coding agents, and... description: Koray Kavukcuoglu, VP of Research at Google DeepMind, discusses the rapid iteration from Gemini 3.0 through 3.6 to 3.7 and the shift from building a model to... canonical: https://daily.dev/posts/koray-kavukcuoglu-on-frontier-models-coding-agents-and-building-agi-iaznshlyz twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Koray Kavukcuoglu on frontier models, coding agents, and building AGI | daily.dev og:description: Koray Kavukcuoglu, VP of Research at Google DeepMind, discusses the rapid iteration from Gemini 3.0 through 3.6 to 3.7 and the shift from building a model to... og:url: https://daily.dev/posts/koray-kavukcuoglu-on-frontier-models-coding-agents-and-building-agi-iaznshlyz og:image: https://api.daily.dev/og/posts/IaZnsHlYZ.png og:image:alt: Koray Kavukcuoglu on frontier models, coding agents, and building AGI og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Koray Kavukcuoglu on frontier models, coding agents, and building AGI **[Google for Developers](https://daily.dev/sources/googledevelopers)** · 26 min read · 1 upvotes · 0 comments ## Summary Koray Kavukcuoglu, VP of Research at Google DeepMind, discusses the rapid iteration from Gemini 3.0 through 3.6 to 3.7 and the shift from building a model to building a coding agent. He confirms Gemini 4 is in training as the most ambitious pretraining run yet, says frontier-model status is the team's singular priority, and reflects on DeepMind's history from DQN and Atari through AlphaGo, AlphaFold, and StarCraft. He argues there is no single test for AGI, describes progress as a series of 'stacked sigmoids,' and gives a brief update on Gemini 3.5 Pro, noting Flash models are improving faster while Pro work continues in parallel. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Is Google DeepMind working on Gemini 4? Yes, Gemini 4 is confirmed to be in active development, described by Koray Kavukcuoglu as the most ambitious pre-training run Google DeepMind has done so far. As of the conversation it was progressing well internally, though no release date or specific capabilities were disclosed, and Gemini 3.5 Pro was still being worked on in parallel. _daily.dev keeps developers tracking Gemini model updates as the AI coding landscape keeps shifting._ ### What changed between Gemini 3.0 and Gemini 3.7 for coding tasks? Google DeepMind shifted focus from building a plain model to building an agent, learning how to handle software engineering and tool use rather than just code generation. Gemini 3.6 launched shortly after 3.0, and 3.7 followed about three weeks later, with architectural and agentic-workflow improvements accumulated over roughly a year converging into those releases. _developers choosing coding agents can follow how these agentic capabilities evolve on daily.dev._ ### Does Google DeepMind believe there is a valid test for AGI? No single benchmark exists that Google DeepMind treats as proof of AGI. Koray Kavukcuoglu states there is no test where hitting a specific score confirms AGI has been reached, and frames progress instead as an ongoing process of building trust through real user interactions rather than crossing a fixed threshold. _developers evaluating AI capability claims can follow this AGI debate as it develops on daily.dev._ --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#google-gemini](https://daily.dev/tags/google-gemini), [#reinforcement-learning](https://daily.dev/tags/reinforcement-learning), [#google-deepmind](https://daily.dev/tags/google-deepmind) [View this post on daily.dev](https://daily.dev/posts/koray-kavukcuoglu-on-frontier-models-coding-agents-and-building-agi-iaznshlyz) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Koray Kavukcuoglu on frontier models, coding agents, and building AGI","url":"https://daily.dev/posts/koray-kavukcuoglu-on-frontier-models-coding-agents-and-building-agi-iaznshlyz","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/koray-kavukcuoglu-on-frontier-models-coding-agents-and-building-agi-iaznshlyz"},"datePublished":"2026-09-01T15:20:50.492Z","dateModified":"2026-09-14T06:46:07.845Z","description":"Koray Kavukcuoglu, VP of Research at Google DeepMind, discusses the rapid iteration from Gemini 3.0 through 3.6 to 3.7 and the shift from building a model to...","image":"https://i.ytimg.com/vi/Rrr2gdbvNFU/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/Rrr2gdbvNFU/sddefault.jpg","isAccessibleForFree":true,"articleSection":"Google for Developers","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Google for Developers","logo":"https://media.daily.dev/image/upload/s--jj3nxRoG--/f_auto/v1725035497/logos/googledevelopers","url":"https://daily.dev/sources/googledevelopers"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/koray-kavukcuoglu-on-frontier-models-coding-agents-and-building-agi-iaznshlyz","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,google-gemini,reinforcement-learning,google-deepmind","timeRequired":"PT26M","video":{"@type":"VideoObject","name":"Koray Kavukcuoglu on frontier models, coding agents, and building AGI","description":"Koray Kavukcuoglu, VP of Research at Google DeepMind, discusses the rapid iteration from Gemini 3.0 through 3.6 to 3.7 and the shift from building a model to...","thumbnailUrl":"https://i.ytimg.com/vi/Rrr2gdbvNFU/sddefault.jpg","uploadDate":"2026-09-01T15:20:50.492Z","duration":"PT26M","url":"https://api.daily.dev/r/IaZnsHlYZ","embedUrl":"https://www.youtube.com/embed/Rrr2gdbvNFU"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Google for Developers","item":"https://daily.dev/sources/googledevelopers"},{"@type":"ListItem","position":3,"name":"Koray Kavukcuoglu on frontier models, coding agents, and building AGI"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/koray-kavukcuoglu-on-frontier-models-coding-agents-and-building-agi-iaznshlyz#faq","mainEntity":[{"@type":"Question","name":"Is Google DeepMind working on Gemini 4?","acceptedAnswer":{"@type":"Answer","text":"Yes, Gemini 4 is confirmed to be in active development, described by Koray Kavukcuoglu as the most ambitious pre-training run Google DeepMind has done so far. As of the conversation it was progressing well internally, though no release date or specific capabilities were disclosed, and Gemini 3.5 Pro was still being worked on in parallel. daily.dev keeps developers tracking Gemini model updates as the AI coding landscape keeps shifting."}},{"@type":"Question","name":"What changed between Gemini 3.0 and Gemini 3.7 for coding tasks?","acceptedAnswer":{"@type":"Answer","text":"Google DeepMind shifted focus from building a plain model to building an agent, learning how to handle software engineering and tool use rather than just code generation. Gemini 3.6 launched shortly after 3.0, and 3.7 followed about three weeks later, with architectural and agentic-workflow improvements accumulated over roughly a year converging into those releases. developers choosing coding agents can follow how these agentic capabilities evolve on daily.dev."}},{"@type":"Question","name":"Does Google DeepMind believe there is a valid test for AGI?","acceptedAnswer":{"@type":"Answer","text":"No single benchmark exists that Google DeepMind treats as proof of AGI. Koray Kavukcuoglu states there is no test where hitting a specific score confirms AGI has been reached, and frames progress instead as an ongoing process of building trust through real user interactions rather than crossing a fixed threshold. developers evaluating AI capability claims can follow this AGI debate as it develops on daily.dev."}}]} ```