Google is quietly testing a new generation of Gemini 3 Pro GA AI models, and leaked details suggest the company is experimenting with multiple internal variants — each built with a very different goal in mind. According to a breakdown from Universe of AI, these prototypes focus not just on raw power, but on safety, speed, cost efficiency, and highly specialized graphics performance. Interestingly, one top-performing version has already been removed from testing, raising questions about how Google balances innovation with responsibility.
At the center of the leak are five codenamed variants: Riftrunner, Snowplow, Snowball, Fire Falcon, and Nova Quida. Together, they represent Google’s attempt to fine-tune AI for different real-world priorities rather than pushing a single one-size-fits-all model.
The Five Gemini 3 Pro Variants Explained
Each prototype appears to serve a specific purpose:
- Riftrunner acts as the baseline reference model used for comparisons
- Snowplow and Snowball are believed to focus on safety and reliability, possibly using different training safeguards
- Fire Falcon and Nova Quida are labeled “Flash” variants optimized for speed and lower operating costs
The Flash versions are still early in testing, but they hint at Google’s interest in creating lighter, faster AI models that could scale more easily across consumer products.
Why SVG Graphics Matter So Much
A major focus of Gemini 3 Pro testing is SVG generation — the ability to produce scalable vector graphics with mathematical precision. SVG images don’t lose quality when resized, making them essential for design, animation, and data visualization.
Leaked demonstrations show the models recreating complex images like the Mona Lisa and generating advanced 3D vector graphics. That level of accuracy suggests Gemini 3 Pro could become a powerful creative tool. However, prioritizing SVG performance may reduce resources available for broader tasks, forcing Google to balance specialization with general intelligence.
The Mystery of the Removed Variant
One variant reportedly outperformed others across multiple benchmarks — yet Google pulled it from testing. While the company hasn’t explained why, analysts speculate the reasons could include:
- Safety or ethical alignment concerns
- Cost inefficiency at scale
- Internal testing failures
- Risk of misuse
The move highlights how modern AI development isn’t just about performance. Companies are increasingly cautious about releasing systems that could create unintended consequences.
A/B Testing Behind the Scenes
Google is using heavy A/B testing to compare these models in live scenarios. This method helps engineers measure reliability, speed, and real-world behavior before any public release. Weaknesses can be identified early, allowing the company to refine or abandon variants before deployment.
Competing With GPT and Anthropic
Against rivals like OpenAI’s GPT 5.3 Codeex and Anthropic’s Cloud Opus 4.6, Gemini 3 Pro appears to stand out in precision graphics and task-specific performance. Whether that advantage holds will depend on Google’s ability to maintain safety, adaptability, and optimization at scale.
Innovation vs. Responsibility
The Gemini 3 Pro leak reveals a company walking a tightrope: pushing AI forward while trying not to outrun its own safeguards. If successful, these models could power creative tools, analytics systems, and next-generation assistants. But Google’s willingness to remove even a high-performing variant shows that caution now plays as big a role as capability.
And in today’s AI race, that balance may matter more than speed alone.
Also Read
Samsung Galaxy S26 Ultra Leak: Privacy Screen, New Design & Camera Upgrades Revealed
Apple Is Killing the Notch: M6 OLED MacBook Pro Leak Changes Everything
Bitmo Lab MeetSticker Review: The Paper-Thin Tracker for “Untrackable” Gear
