Google’s Gemini 3.2 Flash Leaks – Coding Power That Beats Gemini Pro

Gemini 3.2 Flash was silently released on the Gemini web interface before the I/O conference, first spotted by a Reddit user who noticed a drastic change in code style when using the Fast + Canvas mode.

In the new mode the model generates dramatically larger code snippets—up to 2,200 lines of Three.js, SVG, or even a fully interactive Windows 98 environment—from a single prompt, surpassing the previous 400‑500‑line limit of the Flash model.

Developers confirmed the backend change by finding the model entry gemini-3.2-flash-lite-live-preview in the Google Cloud Console, and many reported that selecting “Thinking+Canvas” reliably routes queries to the new model.

Benchmark leaks suggest Gemini 3.2 Flash reaches about 92 % of the performance of the rumored GPT‑5.5 on core coding and reasoning tasks while cutting inference cost by 15‑20× and keeping latency under 200 ms.

The performance boost is attributed to DeepMind’s “model distillation and sparsification” pipeline, which compresses the LLM without the usual trade‑off of accuracy loss.

Beyond coding, Gemini App now integrates third‑party services such as Canva, Instacart, OpenTable, Spotify and WhatsApp, allowing users to issue natural‑language commands like “design a vintage wedding invitation in Canva” or “add the ingredients of this recipe to my Instacart cart.”

These integrations position Gemini as a unified AI assistant that can handle design, shopping, reservations and more without opening separate apps.

Analysts note that while the new model narrows the gap with OpenAI’s GPT‑5.5 and Anthropic’s Claude Mythos, Google still trails the leading competitors in raw model strength, making the upcoming I/O 2026 a critical showdown for the company’s AI leadership.

References: 9to5Google article (https://9to5google.com/2026/05/17/gemini-app-thinking-level/), X post by marmaduke091 (https://x.com/marmaduke091/status/2056052380278374830?s=20).