Google flipped the table! Gemini 3.5 'Snow Bunny' full specs leaked Snow Bunny Checkpoint: Internal leaked version, supports generating complete applications with a single prompt. 3,000 lines of code capability: A single prompt can generate 3,000 lines of production-level runnable code. Fierce Falcon model: Proprietary sub-model focused on extreme response speed and logical processing. Ghost Falcon model: Sub-model specialized in UI design, visual imagery, and native audio creation. Performance surpasses: Benchmark tests outperform GPT-5.2 (75.40%) and Claude Opus 4.5. Deep Think mode: Added manual toggle for 'Deep Think' mode, specifically designed for ultra-complex logic problems. System 2 reasoning: Introduces 'System 2' thinking mechanism, pausing, thinking, and self-verifying before output. 80% reasoning score: Achieved 80% in hardcore reasoning benchmark tests, far exceeding competitors (around 55%). API confirmed: Google codebase already includes the gemini-for-google-3.5 variable, hinting that the interface is ready.