Nano Banana 2.1 is real and generally available: Google's Gemini 3.6 Flash–based image model launched October 6, 2026, and costs less per im

What we checked

• Can the October 2026 model-card publication date and release details be independently confirmed against a current official announcement? confirmed. The DeepMind model card is dated October 6, 2026 and says the model is based on Gemini 3.6 Flash. Google's deprecations table lists gemini-nano-banana-2.1 with an October 6, 2026 release date, and the models page lists it as 'New Stable'. The Decoder and The Next Web independently report the same date and base model. I did not find a blog.google announcement post. [1][4][5][6][7]
• (Added) Is Nano Banana 2 (gemini-3.1-flash-image) scheduled to shut down on October 29, 2026? refuted. Google's deprecations page lists gemini-3.1-flash-image with 'No shutdown date announced' and names gemini-nano-banana-2.1 as its replacement. The October 29 date reported in some third-party coverage does not appear on that page. The page does say that shutdown dates are announced in advance. [4][7]
• (Added) Is the 4K image price about 7.56 cents, as The Decoder reports? refuted. Google's pricing page lists $0.113 per 4K image. That is 3,780 output tokens at $30 per million image-output tokens. The figure of about 7.56 cents is the batch price for Nano Banana 2 at 4K, or the token count of the old model applied to the new rate, not the standard price for 2.1. [2][6]

What we found

• The model card is dated October 6, 2026 and states 'Nano Banana 2.1 is based on Gemini 3.6 Flash.' It accepts text and image inputs (it cites a context window of up to 1M tokens) and produces images plus text output of up to 64K tokens. It lists availability in the Gemini app, AI Studio, Gemini API, AI Mode in Search, Google Ads, Flow and Stitch. One oddity: the frontier-safety section refers to evaluations of Gemini 3.1 Pro and 'Gemini 3.7 Flash' rather than 3.6 Flash. [1]
• Google's own human-preference Elo scores (thinking mode) put 2.1 at 1050 for overall text-to-image preference, against 990 for Nano Banana 2 and 935 for Nano Banana Pro. Multi-character consistency scores are 1106, 978 and 1011, and the infographic factuality autorater gives 0.521, 0.179 and 0.265. Without thinking, overall preference drops to 1015. These are vendor benchmarks only. [1]
• Official standard pricing for gemini-nano-banana-2.1 is $1.50 per million input tokens, $7.50 per million text/thinking output tokens and $30 per million image output tokens. Per image, that works out to $0.0336 at 1K, $0.0504 at 2K and $0.113 at 4K, with no free tier. Batch pricing is about half. Nano Banana 2 costs $0.067, $0.101 and $0.151 per image, but its input is $0.50 and its text output $3.00 per million tokens. So 2.1 is cheaper per image but about 3x more expensive for input. Also, 4K images now use 3,780 tokens instead of 2,520. [2]
• The Gemini API image-generation docs confirm the model ID gemini-nano-banana-2.1. They list thinking levels of minimal, medium (the default) and high, and up to 14 reference images (the extracted table suggests 10 objects and 4 characters). They also confirm Google Search grounding including Image Search, 1K/2K/4K resolutions, and SynthID watermarks on all outputs. The 512px tier is not offered on 2.1. [3]
• The deprecations page lists gemini-3.1-flash-image (released May 28, 2026) with 'No shutdown date announced' and gemini-nano-banana-2.1 as its replacement. The models page now calls gemini-3.6-flash a 'previous-generation' Flash model and lists a newer gemini-3.8-flash. In other words, the image model is built on a text base that has already been superseded. [4][5]
• Independent coverage confirms the date and the base model but adds caveats. In The Decoder's hands-on test, Nano Banana Pro often produced more realistic images, while 2.1 got scale wrong. The same article also gives a 4K price of 7.56 cents, which conflicts with Google's pricing page. The Next Web confirms the price change and notes that free Gemini app users were moved to Flash-Lite on October 9. [6][7]

What it means

The release is genuine, the dates line up, and builders can treat gemini-nano-banana-2.1 as a stable production option. Before switching, recalculate your costs from Google's pricing page rather than from press tables. Workloads that mostly generate images will usually get cheaper at 1K and 2K. Workloads with long prompts, many reference images or a lot of thinking and text output may cost less than expected, or even more, because input and text-output rates went up. The default 'medium' thinking level adds billed thinking tokens. Benchmark minimal and medium against your own prompts before you set a default.

There is no announced shutdown for Nano Banana 2, so you have time to migrate. Use it to run side-by-side quality tests against both Nano Banana 2 and Pro, because independent testing found Pro still wins on realism in some cases. Expect known weak spots: small or dense text, left/right confusion, imperfect character consistency, and partial instruction-following in mask-based editing. Plan for SynthID watermarks on every output.

Still open

• What are the exact API input and output token limits? The model card cites a 1M-token context and 64K text output, but I could not see the API limits in the docs I opened.
• Will Google publish a shutdown date for gemini-3.1-flash-image, and when?
• Why does the model card's frontier-safety section refer to Gemini 3.7 Flash when the model is based on Gemini 3.6 Flash?
• Is there an official blog.google announcement, and does it add any details? I did not locate one.