A series of leaks has surfaced, pointing to Google's plans to launch two cutting-edge AI models—Gemini 3 Pro and Nano Banana 2—before the end of 2025. References to these models appeared on Google's Vertex AI cloud platform and the official Gemini website, signaling a significant upgrade to the tech giant's AI arsenal that could reshape developer workflows and intensify the generative AI race.

Gemini 3 Pro: A Coding Powerhouse with Massive Context

The first leak centers on Gemini 3 Pro, spotted under the identifier gemini-3-pro-preview-11-2025 in Vertex AI, Google's hosted platform for building AI applications. This naming convention aligns with earlier reports of a November 2025 rollout. While Google typically tests such models under strict NDAs with select partners, preventing public disclosures, early insights suggest a groundbreaking 1-million-token context window. This leap could enable more complex, long-form coding tasks and reasoning.

Comparisons highlight Gemini 3 Pro's potential: Google's current flagship, Gemini 2.5 Pro, released eight months ago, remains competitive with a 63.8% score on SWE-Bench Verified (a benchmark for agentic coding evaluations) using custom setups. However, it trails Claude Sonnet 4.5's 77%. Gemini 3 Pro aims to close this gap, emphasizing Google's focus on making AI an indispensable tool for developers tackling intricate software engineering challenges.

Nano Banana 2: Revolutionizing Image Generation

The second model, Nano Banana 2 (codenamed GEMPIX2), emerged via references on the Gemini website, hinting at a December 2025 launch. Positioned as a leader in realistic image synthesis, it builds on its predecessor to offer enhanced capabilities for creative and technical applications, from digital art to multimedia content generation. This move underscores Google's ambition to dominate not just text-based AI but also the rapidly evolving visual AI landscape.

Industry Implications and Competitive Pressures

Google isn't alone in this sprint—OpenAI is reportedly developing GPT-5.1 and new Codex models for its agentic coding terminal. This back-to-back innovation cycle highlights an industry-wide push to deliver more specialized, efficient AI tools. For developers, the arrival of Gemini 3 Pro could mean faster prototyping and debugging, while Nano Banana 2 opens doors for seamless integration of AI-generated visuals into apps. Yet, it also raises questions about resource demands and the need for robust security practices as these models handle sensitive data.

As the AI field accelerates, leaks like this serve as a reminder that transparency often lags behind progress. The true test will come when developers get their hands on these tools, evaluating whether they deliver on promises to enhance productivity without compromising reliability.

Source: BleepingComputer