Given how solid and remarkable Gemini 2.5 has been so far, expectations are running high from Gemini 3. Well, Google is today releasing its new AI model and it seems that Gemini 3 is indeed smart and good enough to beat the current state-of-the-art AI models such as ChatGPT 5 and Grok 4. Announcing Gemini 3, the company describes its latest AI as its most advanced system so far, capable of interpreting information with a level of depth and nuance that is closer to human reasoning.
Google says it is rolling out Gemini 3 to users across its platforms, including on Search. It is also going to be available in the Gemini app to all users, although there will be different usage limits on it depending on which Google AI plan you have subscribed to. Google says that currently over 650 million users are using Gemini AI every month.
According to CEO Sundar Pichai, Gemini 3 brings together Google’s work in multimodal understanding, long-context analysis and agentic behaviour in one AI system. “Gemini 3 is state-of-the-art in reasoning, built to grasp depth and nuance — whether it’s perceiving the subtle clues in a creative idea, or peeling apart the overlapping layers of a difficult problem,” the Google CEO noted in a blog. “Gemini 3 is also much better at figuring out the context and intent behind your request, so you get what you need with less prompting. It’s amazing to think that in just two years, Al has evolved from simply reading text and images to reading the room.”
Google says Gemini 3 is built for reasoning
Demis Hassabis, CEO of Google DeepMind, explained the AI system to users in another note. “Today we’re taking another big step on the path toward AGI and releasing Gemini 3,” wrote Hassabis. “This means Gemini 3 Pro is highly capable at solving complex problems across a vast array of topics like science and mathematics with a high-degree of reliability. Gemini 3 Pro also brings a new level of depth and nuance to every interaction. Its responses are smart, concise and direct, trading cliche and flattery for genuine insight.”
Google claims Gemini 3 performs strongly on academic and synthetic tests that measure analytical depth, including Humanity’s Last Exam, GPQA Diamond, and a new high score on MathArena Apex, which examines mathematical and symbolic problem-solving.
Capabilities of Gemini 3
Google says the new AI model can analyse and understand different types of content. This includes handwritten notes, long research papers and even video lectures. According to the company, users might apply these abilities in scenarios such as converting family recipes into digital formats, creating structured notes from dense academic material, or breaking down long, complex videos into step-by-step explanations.
The model will also underpin new features in AI Mode on Google Search, where it can generate dynamic visual layouts, simulations and interactive tool-based results.
Gemini 3 Deep Think is in experiment
Alongside the main model, Google has also previewed Gemini 3 Deep Think, an extended-reasoning mode designed for demanding analytical tasks. In internal testing, Google says it performs slightly better than Gemini 3 Pro on higher-difficulty benchmarks and shows stronger tool-use abilities, particularly when running code or validating hypotheses.
Although, for now the Deep Think mode is not going to be available to users. Instead, Google says that this mode is going to be tested by researchers and hand-picked testers for now. Once the company is satisfied that the Think Mode can be released safely, it will likely rollout to Gemini Ultra subscribers.
Gemini 3 brings Antigravity for developers
Building on top of the new AI system, Google is also expanding the agentic and coding features of Gemini, particularly for developers.
For developers, Gemini 3 apparently comes with stronger coding performance and more consistent agentic behaviour, the ability for the model to plan tasks, use tools and carry out multi-step actions over longer periods. According to Google, the Gemini 3 currently leads the WebDev Arena leaderboard and shows improved results on coding-agent tests such as SWE-bench Verified and Terminal-Bench 2.0.
Google is also launching Google Antigravity, a new software development tool where AI agents can operate more autonomously across an editor, terminal and browser. Instead of acting purely as a suggestion tool, the company says that the AI agent can plan and execute entire workflows, almost like a junior developer, while the user remains in control.
