Google DeepMind is signaling that its next flagship artificial intelligence model, Gemini 4, is close to release, as the company tries to reclaim momentum in a race that has lately been defined by rivals OpenAI, Anthropic, and a fast-moving open-source ecosystem. In his first media appearance as the new chief of Google DeepMind, Koray Kavukcuoglu told The Information that Gemini 4 is in its refinement stage and that the company aims to launch it 'much earlier' than the end of the year.

The timing matters. Google has faced persistent questions about whether it is lagging on flagship AI releases, even as it ships smaller models and integrates Gemini across Search, Workspace, Android, and Cloud. Kavukcuoglu’s comments, reported by The Verge, suggest Google wants to move from defense to offense—and to do so before competitors set the narrative for the next generation of AI.

Our intention is to, like - as soon as possible - to release an early post-training output because we see the results and we are excited.

Kavukcuoglu added that Google would 'continue the fast-paced iteration' that has characterized its recent AI work. That phrasing is notable: it implies Gemini 4 may not arrive as a single, finished monolith but as a series of post-training updates, evaluations, and product integrations. In practice, that could mean developers and enterprises see early versions through APIs and cloud services before a broader consumer launch.

A missing model and a crowded roadmap

The Gemini 4 talk comes amid confusion about the company’s naming and release schedule. One report aggregated by MSN asks, 'Where is Gemini 3.5 Pro? The case of the missing AI model.' The question reflects a broader sense in the developer community that Google has skipped, delayed, or quietly shelved certain intermediate versions while it reallocates compute and talent toward more strategic releases.

At the same time, another MSN report says Google is 'almost ready to launch new Gemini AI model' and may beat Anthropic and OpenAI in coding this time. That claim is significant because coding assistants have become a key battleground for AI labs. Developers judge models on benchmarks such as SWE-bench, HumanEval, and real-world repository tasks, and coding tools are among the fastest paths to revenue and enterprise adoption.

Separately, a report highlighted by MSN says the Gemini 3.8 Flash AI model has 'significantly upgraded coding capabilities.' If accurate, that suggests Google is not waiting for Gemini 4 to improve its developer offerings. Flash models are typically optimized for speed and cost, so stronger coding performance there could matter as much as a flagship launch—especially for startups and enterprises that need low-latency agents and code review tools.

From chatbot to agent

Google DeepMind is also reframing what Gemini is supposed to be. Search Engine Journal reports that DeepMind says Gemini is evolving from chatbot to AI agent. That shift is not merely semantic. A chatbot answers questions; an agent plans, uses tools, remembers context, and acts across software. For Google, agents could be woven into Gmail, Docs, Chrome, Android, and Cloud, turning Gemini from a destination into an operating layer.

The agent vision also explains why Google is talking about iteration rather than a single launch day. Agentic systems require post-training, reinforcement learning, safety evaluations, and integration with external tools. They also raise new risks: prompt injection, data leakage, unintended actions, and accountability when an autonomous system makes a mistake. Google’s willingness to release an 'early post-training output' suggests it is balancing competitive pressure against the need for real-world feedback.

Robots and the embodied AI bet

Perhaps the most ambitious thread comes from Scientific American, which reports that Google DeepMind wants to build an AI brain that can jump between robot bodies. The idea is to use Gemini as a foundation model for many different robots, rather than training a separate system for each hardware platform. That would mirror the way large language models generalize across text tasks: one model, many embodiments.

If successful, such a system could accelerate robotics by letting developers fine-tune a shared Gemini-based brain for warehouses, factories, homes, or labs. It would also intensify competition with Nvidia, Tesla, Figure, and other players pursuing embodied AI. But the technical hurdles are steep. Robots must perceive in 3D, act in real time, and cope with messy physical environments. A model that works in simulation may fail on a factory floor. Google’s advantage is its combination of AI research, cloud infrastructure, and a growing robotics portfolio, but execution will be difficult.

Inside the engine room

Culture and organization matter, too. CyberNews quotes Google DeepMind’s head of community, Amit Vadi, saying, 'We’re the engine room of Google AI.' The line captures how DeepMind has become the central research and product engine for Google’s AI ambitions, especially after the company merged its Brain and DeepMind teams. That consolidation gave DeepMind more control over models, infrastructure, and talent—but it also raised expectations.

Different outlets are framing the story in different ways. The Verge focuses on the launch timeline and Kavukcuoglu’s first interview. MSN’s aggregation highlights the missing 3.5 Pro and the coding race. Search Engine Journal emphasizes the chatbot-to-agent transition. Scientific American looks at the robotics frontier. CyberNews offers an insider view of DeepMind’s self-image. Together, they paint a picture of a company trying to move on several fronts at once: flagship model, developer tools, agents, and embodied AI.

Why it matters

The stakes are high for Google, for developers, and for the broader AI market. If Gemini 4 lands early and performs well in coding and reasoning, it could reset perceptions that Google is perpetually catching up. If it slips, rivals will have more time to lock in enterprise customers, developer mindshare, and ecosystem lock-in. The rise of agents and robotics also raises questions about safety, labor, and regulation that go beyond benchmark scores.

For now, the clearest signal is that Google DeepMind is not waiting for a perfect end-of-year moment. Kavukcuoglu’s comments suggest a company eager to show results, gather feedback, and iterate in public. Whether that strategy produces a true Gemini 4 breakthrough—or simply more incremental updates—will depend on how quickly Google can turn research excitement into reliable products. The AI race is no longer just about who has the best chatbot. It is about who can build the most capable, trustworthy, and widely deployed AI agents—and eventually, the brains that power the physical world.