Google has officially launched Gemini 3 Pro, its next-generation flagship AI model, on November 18, 2025, making it available in preview for developers through the Gemini API. The announcement marks a significant leap forward in AI capabilities, with Google claiming a more than 50% improvement in reasoning and a 35% increase in accuracy on software engineering tasks compared to its predecessor, Gemini 2.5 Pro. This release is not just an incremental update; it introduces a new architecture and a suite of powerful tools designed to enable more complex and autonomous AI applications.
What happened with the Gemini 3 Pro launch?
On November 18, 2025, Google made Gemini 3 Pro available to developers via the Gemini API in Google AI Studio and Vertex AI. The launch introduces the most intelligent and capable model from Google to date, setting new state-of-the-art performance across a wide range of benchmarks, including multimodal understanding, coding, and agentic workflows. Alongside the core model, Google unveiled “Nano Banana Pro” (the official model name is `gemini-3-pro-image-preview`), an advanced image generation and editing tool built on Gemini 3 Pro. Additionally, Google launched Google Antigravity, a new agentic development platform designed to showcase the model’s ability to handle complex, multi-step software development tasks autonomously.
Why it matters: A new toolkit for developers
For developers, the Gemini 3 Pro launch is significant for four key reasons: enhanced reasoning, superior multimodal capabilities with granular control, powerful new agentic features, and state-of-the-art image generation.
- Advanced Reasoning and API Control: The core of Gemini 3 Pro is its enhanced reasoning engine. To give developers control over this power, the API introduces a new
thinking_levelparameter. This allows developers to balance reasoning depth against latency and cost, choosing “high” for complex problem-solving or “low” for faster, simpler tasks. - Superior Multimodality: Gemini 3 Pro excels at understanding and processing combined inputs of text, images, audio, and video. The API now includes a
media_resolutionparameter, giving developers granular control over the token cost and fidelity of visual processing, which is critical for applications involving dense document OCR or fine-grained video analysis. - Agentic Capabilities & Thought Signatures: The model is designed for agentic workflows, where it can plan and execute multi-step tasks. A crucial new technical feature is Thought Signatures (
thoughtSignature), an encrypted token that preserves the model’s reasoning context across multiple API calls. Developers must pass this signature back in subsequent requests to maintain the model’s train of thought, which is strictly enforced for function calling and image editing. - Nano Banana Pro (Gemini 3 Pro Image): This new tool, powered by Gemini 3 Pro’s reasoning, brings grounded, conversational image generation. It can use Google Search to verify facts before generating an image (e.g., creating an accurate weather infographic), render sharp text, produce images up to 4K resolution, and perform multi-turn edits based on simple conversational commands.
Impact and implications for the future
The release of Gemini 3 Pro signals a shift from instruction-following AI to problem-solving AI. For developers, this means the ability to build applications that can autonomously tackle complex, long-horizon tasks—from planning and executing a software feature across an entire codebase to synthesizing information from multiple documents and generating a data visualization. The introduction of platforms like Google Antigravity further suggests that the future of development will involve collaborating with AI agents that manage their own workflows. However, developers must adapt to new API requirements, such as managing Thought Signatures, to fully leverage the model’s power. As of November 2025, Gemini 3 Pro is available in preview, and developers can start exploring its capabilities in Google AI Studio.
