Google DeepMind chief Koray Kavukcuoglu stated that maintaining frontier AI leadership is the company’s single most important priority, pushing back against claims that Google prioritizes price-to-performance over model capability. While acknowledging that Google’s current AI models sit slightly below the frontier, Kavukcuoglu emphasized that the organization possesses the necessary team, resources, and full-stack capabilities to close the gap.
Kavukcuoglu declined to share new updates on upcoming releases, confirming only that Gemini 4 remains their most ambitious training run to date and that Gemini 3.5 Pro remains in development despite delays. Instead, he highlighted rapid advancements in the Gemini Flash series, framed as a structural evolution from traditional language models into autonomous coding agents.
According to Kavukcuoglu, Google DeepMind is strategically targeting software engineering as the primary environment for agentic systems. Moving forward, the company’s engineering efforts aim to transform its core architectures into agents capable of executing complex domain-specific tasks.
Why it matters
Signals Google DeepMind’s strategic pivot toward autonomous software engineering agents as the core domain for frontier capabilities.
Confirms Gemini 3.5 Pro is delayed, indicating technical or compute bottlenecks in Google’s intermediate model pipeline.
Reassures investors and enterprise clients that Google remains committed to high-end frontier research despite cost-efficiency market pressures.
Source: the-decoder.com
