Google Cuts AI Image Prices in Half: Nano Banana 2.1 Generates a 1K Image for $0.0336 as the Old Model Gets an October 29 Shutdown Date

Google has shipped Nano Banana 2.1, an updated image generation and conversational editing model that is live in the Gemini app, AI Mode in Search, Google AI Studio, Flow, Stitch, Google Ads and the Gemini Enterprise platform. In the Gemini API it is available as gemini-nano-banana-2.1, and it arrives with the sharpest change developers will feel: image output prices have been cut roughly in half.

At the new rates, a 1K image costs about $0.0336, down from roughly $0.067 for Nano Banana 2; 2K output falls from about $0.101 to $0.0504; and 4K generation is listed at $0.0756 per image in early coverage of the API changelog, with batch requests halving those prices again. The underlying mechanism is a cut in the image output token price from $60 to $30 per million — though input tokens rose from $0.50 to $1.50 per million and text-and-thinking output climbed to $7.50, so teams sending long prompts or many reference images may see a smaller net saving than the per-image figures suggest. All figures are company-published rates, not independently audited benchmarks.

The capability list is aimed squarely at production creative work. The model accepts up to 14 reference images per request and can hold up to four characters and ten objects consistent across multi-turn edits, supports extreme aspect ratios including 1:4, 4:1, 1:8 and 8:1 at 1K, 2K and 4K resolutions, and improves mask-based editing and subject consistency. Text rendering — the traditional weak spot — got a specific claim: Google says its infographic factuality score rose from 0.179 to 0.521. New features include grounding in Google Search and Google Image Search, three adjustable thinking levels, and C2PA Content Credentials for provenance tracking.

Developers on the previous model have a hard deadline. Google has set October 29, 2026 as the earliest shutdown date for gemini-3.1-flash-image, the API model behind Nano Banana 2, leaving teams roughly three weeks to migrate and re-test prompts that depend on text rendering or multi-character edits.

One oddity worth flagging: Google's own documentation does not agree on what the new model is built on. The official model card published by Google DeepMind says Nano Banana 2.1 is based on Gemini 3.6 Flash, while the Gemini API documentation describes it as an update to Gemini 3.1 Flash Image. Google has not published a reconciliation. Its model card also concedes weaker performance on spatial localisation, world knowledge, 3D reasoning and general factuality, and independent testers have flagged small-text rendering and occasional "pose leakage" in edited subjects.

Independent leaderboards add context to Google's "outperforms our previous models across the board" framing. On the Arena text-to-image leaderboard, Nano Banana 2.1 currently sits fifth behind several versions of OpenAI's GPT Image 2.5, and previous testing found the flagship Nano Banana Pro still produces more realistic images in practice — 2.1 is priced as the fast, cheap workhorse, not the quality ceiling.

The strategic signal is a price war in image generation. Google, OpenAI and a wave of open-weight models from Chinese labs are all competing for developers who embed image features into apps, and halving the per-image cost pushes AI-generated visuals further into work that used to require a designer for every revision — localized banners, product catalogs, menu cards and simple infographics. Teams should still benchmark with their own prompts before assuming their bill will halve, but the direction of the market is unambiguous: down.