Google released Nano Banana 2.1 on 6 October, an update to its Gemini image model that the company says improves visual quality, multi-turn character consistency, text rendering and search-grounded generation — while cutting the price of the images it produces roughly in half.
The model is built on Gemini 3.6 Flash and is documented as a member of the Gemini 3 series of natively multimodal reasoning models. On the Gemini API, paid-tier image output is listed at $30 per million tokens, which Google equates to $0.0336 for a 1K image, $0.0504 for 2K and $0.0756 for 4K. The model it replaces, Gemini 3.1 Flash Image — known as Nano Banana 2 — was priced at $60 per million image tokens, or $0.067, $0.101 and $0.151 per image at the same three resolutions. Text and thinking output costs $7.50 per million tokens, and text, image and video input $1.50 per million. Batch jobs run at half those rates.
Google deprecated Nano Banana 2 the same day and lists a shutdown date of 29 October 2026, leaving developers a little over three weeks to move to the new identifier, gemini-nano-banana-2.1.
The published limits are broad: text and images as input, image and text as output, video input only, and no audio. Enterprise documentation lists a 131,072-token context window with up to 32,768 output tokens, support for up to 14 images per prompt in ratios from 1:1 to 21:9, and 1K, 2K and 4K output. Each input image consumes 1,120 tokens; so does a 1K output image, while a 2K output image consumes 1,680. The model can process up to 14 reference images at once while keeping up to four characters and ten objects consistent, and offers minimal, medium and high thinking levels that trade latency for image quality. Multi-turn editing, interleaved image and text, C2PA Content Credentials and virtual try-on are supported; generating people is not.
The published numbers need reading with care. Google's model card reports human-evaluation Elo scores above both Nano Banana 2 and Nano Banana Pro across text-to-image and editing tasks — an overall preference Elo of 1050 ±14 in the thinking configuration for text-to-image, and 1106 for multi-character consistency in editing — and states that the model satisfied required child-safety launch thresholds. But in that same table the model being retired, Nano Banana 2, also outscores Gemini 3 Pro Image everywhere, and side-by-side tests found Nano Banana Pro still producing the more realistic images in practice, with 2.1 struggling on scale. Google describes 2.1 as the more efficient counterpart to Pro. The card also lists the usual caveats: hallucinations, occasional slowness or timeouts, poor rendering of small text and long paragraphs, imperfect character consistency between input and generated images, and limited world knowledge, 3D reasoning and factuality.
Availability spans the Gemini app, Google AI Studio, the Gemini API, AI Mode in Google Search, Google Ads, Google Flow and Google Stitch. Google lists no free tier for the new model.
