Leaks Confirm Google Gemini 3 Pro and Nano Banana 2 Could Launch Soon

Google is preparing to launch its next major AI models, Gemini 3 Pro and Nano Banana 2, according to new leaks.

On November 7, a preview model named “gemini-3-pro-preview-11-2025” was found on Google’s Vertex AI platform, pointing to a November release. At the same time, code for a “Nano Banana 2” image generator surfaced on the main Gemini website, according to user posts on X.

These developments show Google is moving fast to challenge new AI tools from rivals OpenAI and Anthropic. The company aims to win over developers with stronger coding and image creation abilities.

A Trail of Leaks: From Stealth Tests to Platform Spoilers

Spotted first on Google’s own Vertex AI platform, the new model’s name explicitly suggests a November 2025 preview window. Multiple reports confirm the finding and highlight a significant technical upgrade: leaks suggest Gemini 3 Pro will feature a 1 million token context window.

The  large context window would allow the model to process and reason over entire codebases, lengthy financial reports, or hours of transcribed video in a single prompt. It would unlock more sophisticated, long-form analysis and content generation that current models struggle with.

A developer-focused analysis also noted a potential ‘thinking’ variant of the model, hinting at different operational modes for tackling complex problems.

According to user reports, some chinese api aggregators are already offering access to Gemini 3.0.

Concurrently, code referencing “Nano Banana 2,” codenamed “GEMPIX2,” reportedly appeared on the Gemini website, signaling an imminent update to the company’s popular AI image generator.

This latest evidence follows a month of intense speculation, fueled by Google’s own stealthy rollout strategy.

The company began quietly A/B testing Gemini 3.0 with select developers in early October before providing the model to some users without any official announcement around October 15.

Public intrigue peaked when two mysterious but powerful models, codenamed ‘lithiumflow’ and ‘orionmist’, appeared on the public testing platform LMArena. Testers reported exceptional performance in tasks like generating complex SVG graphics from text prompts, further fueling beliefs that these were pre-release versions of Gemini 3.0.

While Google remains officially silent on a launch date, CEO Sundar Pichai previously confirmed a new version would arrive “later this year”.

This gradual, data-driven release preparation marks a strategic pivot from the splashy launch events common in the tech industry, allowing Google to fine-tune performance based on real-world usage.

Racing Against a New Generation of AI Coders

Google’s push comes at a critical moment in the AI arms race. The competitive field has escalated rapidly, with rivals setting an incredibly high bar in just the past month.

This puts immense pressure on Google to deliver a state-of-the-art product that can compete on both raw performance and new, long-duration agentic capabilities.

In mid-September, OpenAI launched GPT-5-Codex. Its key innovation is “dynamic thinking,” which allows the model to adjust its computational effort in real-time, effectively deciding to “think harder” about a complex problem.

Just two weeks later, Anthropic fired back with Claude Sonnet 4.5. Anthropic highlighted its model’s extraordinary endurance, stating it can operate autonomously on software projects for over 30 hours.

The capabilities of these new models have been praised by partners. These advancements mean Gemini 3.0 Pro must not only match but exceed these benchmarks to reclaim a leadership position.

Powering Google’s Broader Enterprise Ambitions

Beyond the model’s raw power, the release of Gemini 3.0 Pro is a cornerstone of Google’s broader enterprise strategy. The new model is expected to power Gemini Enterprise, a comprehensive platform for businesses to build and deploy custom AI agents.

Google Cloud CEO Thomas Kurian framed this as a key differentiator, arguing that unlike rivals, Google provides a unified, end-to-end solution.

In a clear jab at competitors, he said, “some companies offer AI models and toolkits, but they are handing you the pieces, not the platform. They leave your teams to stitch everything together.”

This “toolkit” approach often requires businesses to have significant in-house AI expertise to integrate different models, APIs, and safety filters. Google’s pitch is a pre-integrated platform where the model, development environment like Vertex AI, and deployment infrastructure work seamlessly, lowering the barrier to entry for companies without dedicated AI research teams.

As Gartner analyst Chirag Dekate noted at the launch of Gemini Enterprise, “how Google is able to leverage this unified messaging in the Gemini 3.0 launch sequence… will also be a crucial litmus test.” For now, the industry watches to see if this quiet rollout will soon become a roar.