Top Architect

Jun 10, 2026 · Artificial Intelligence

Gemini 3.2 Flash Unveiled: How Google’s New Model Outcodes Its Own Pro in Code Generation

Google quietly released Gemini 3.2 Flash on the web, where developers discovered a hidden model that, when triggered via Thinking + Canvas, generates massive, high‑quality code—up to 2 200 lines for complex 3D, Windows 98, and PS5 UI tasks—while delivering 15‑20× lower inference cost, sub‑200 ms latency, and deep app integrations, marking a major AI industry milestone.