Recently, the Gemini series AI model under Google experienced a major leak. According to the latest intelligence, an internal model codenamed "Snow Bunny" was exposed, and its powerful features have caused a stir in the industry. As AIbase editors, we have organized and analyzed this information to provide readers with professional insights. The following is a report compiled from reliable sources, aiming to showcase the potential breakthroughs of Gemini 3.5.

image.png

Exposure of "Snow Bunny" Checkpoint: Build a Complete Application in One Go

The leaked internal model "Snow Bunny" is described as a revolutionary checkpoint that can generate and build an entire application from a single prompt. This feature marks a leap in AI within the software development field, greatly simplifying the process from concept to final product and is expected to reshape the workflow for developers.

Remarkable Code Generation Ability: Output 3,000 Lines of Working Code with a Single Prompt

According to the leaked details, this model has the capability to generate up to 3,000 lines of executable code with just one prompt. This not only improves efficiency but also ensures the practicality and reliability of the code, offering an efficient solution for complex projects.

New Model Lineup Revealed: Fierce Falcon and Ghost Falcon

Gemini 3.5 introduced two specialized models. "Fierce Falcon" focuses on pure speed and logical processing, suitable for scenarios requiring fast computation; "Ghost Falcon" excels in UI design, visual elements, and audio creation, expanding AI's applications in the multimedia field. The combination of these two models enhances the comprehensiveness of Gemini.

Performance Surpasses Competitors: Outperforms Unreleased GPT-5.2 and Claude Opus 4.5

In benchmark tests, Gemini 3.5 performed outstandingly. It scored 75.40%, surpassing the unreleased GPT-5.2 and outperforming Claude Opus 4.5. This comparison highlights Google's leading advantage in AI performance optimization.

Deep Thinking Mode Activated: Solve Complex Logical Problems

The new feature includes the "Deep Think" switching mode, designed specifically for handling high-level logical problems. This mode allows the AI to conduct in-depth analysis before responding, improving the accuracy of problem-solving.

System2 Reasoning Mechanism: Pause Before Thinking and Then Output

Gemini 3.5 uses the "System2" thinking framework, which actively pauses and reasons before answering. This human-like cognitive mechanism ensures deeper and more reliable outputs.

Leading Reasoning Benchmark Score: 80% vs. Competitors' 55%

In difficult reasoning benchmark tests, the model achieved a score of 80%, far exceeding competitors' 55%. This data proves Gemini's strong competitiveness in complex tasks.

API Interface Confirmed: gemini-for-google-3.5 Variable Ready