Google: Nano Banana 2.1

Gemini Nano Banana 2.1 (API ID gemini-nano-banana-2.1) is Google DeepMind’s October 2026 refresh of the Flash-tier Nano Banana image stack, built on the Gemini 3.6 Flash backbone within the broader Gemini 3 family. It targets high-volume text-to-image generation and conversational editing with 1K, 2K, and 4K output, stronger text rendering and infographic layout, multi-turn character consistency, mask and ink-style edits, and fusion of up to 14 reference images. Inputs include text, images, video, and PDFs via the Gemini API; outputs are images and text. Configurable thinking levels (minimal, medium, high) and grounding with Google Web and Image Search support fact-aware and visually grounded creation.

Google’s model card reports substantial gains over Nano Banana 2 (Gemini 3.1 Flash Image) and Nano Banana Pro on internal human-preference and editing Elo suites—for example 1050 overall text-to-image preference and 1106 multi-character consistency in Thinking mode—while keeping roughly half the per-image cost of Nano Banana 2 at shared resolutions. Independent LMSYS Arena leaderboards (October 2026, preliminary) place it among the top Google image models, with competitive text-to-image and editing scores though still behind leading OpenAI GPT Image variants on those boards.

The model is distributed only through Google’s cloud and consumer surfaces (Gemini API, AI Studio, Gemini App, Vertex/Gemini Enterprise, and selected Google products). It is not open-weight; weights and training details follow the proprietary Gemini 3.6 Flash lineage. Known limitations include imperfect small-text rendering, occasional spatial or instruction-following errors in edits, and residual factuality gaps despite search grounding.