Nano Banana 2.1

Changes from previous version

First Nano Banana version in this catalog. Image output is $30 per million tokens, half the $60 rate of Nano Banana 2, and the 512px tier is gone. Thinking-mode overall preference is 1050 versus 990, infographic factuality is 0.521 versus 0.179, and mask-based editing is 1049 versus 965.

Release Summary

Google's high-efficiency image model after Nano Banana 2 (Gemini 3.1 Flash Image), served as gemini-nano-banana-2.1. Image output is $30 per million tokens: $0.0336 at 1K (1,120 tokens), $0.0504 at 2K (1,680 tokens), and $0.113 at 4K (3,780 tokens). Text and thinking output is $7.50 per million tokens, and input is $1.50. Batch is half those rates. There is no free tier. Thinking levels are minimal, medium (the default), and high, with up to 14 reference images. Google's thinking-mode overall preference is 1050 Elo, plus or minus 14, versus 990 for Nano Banana 2 and 935 for Nano Banana Pro. Infographic factuality is 0.521, versus 0.179 for Nano Banana 2. The Gemini API context window is 131,072 tokens and max output is 32,768. It is based on Gemini 3.6 Flash. The knowledge cutoff is March 2026, and some domains stay limited to January 2025.

Timeline