Google Cuts Nano Banana 2.1 Image Prices in Half While Climbing Rankings

Google has made Nano Banana 2.1 generally available through the Gemini API as gemini-nano-banana-2.1. The release halves image-output prices while improving benchmark rankings, and developers using its predecessor must migrate before Google retires that model on October 29, 2026.

Built on Gemini 3.6 Flash, Nano Banana 2.1 succeeds the deprecated gemini-3.1-flash-image. Nano Banana Pro remains available as the higher-priced tier. Artificial Analysis ranked the new model fourth overall on both its AA-Image-T2I v2.0 and AA-Image-Editing v2.0 leaderboards, three places above Nano Banana 2 on each.

Better images for half the output price

Google bills generated images as output tokens. Nano Banana 2.1 costs $30 per million output tokens, down from $60 for Nano Banana 2.

At 1K resolution, the new rate is half the Nano Banana 2 price and one-quarter of the Nano Banana Pro price. Artificial Analysis reports that no cheaper model scores higher on its text-to-image leaderboard, while the three models above Nano Banana 2.1 cost about six times more per image.

Generation takes about 16 seconds for a 1K image, compared with 8.3 seconds for Nano Banana 2. Every higher-ranked text-to-image model takes at least 3.3 times as long, making Nano Banana 2.1 the fastest model among the leaderboard’s top six.

Structure and identity drive the gains

Nano Banana 2.1 improved across all 16 capabilities measured by Artificial Analysis. Its largest advances address several weaknesses in the previous release:

• Layout: Improved from #10 to #4 on flows, arrows, visual hierarchy, and multi-panel compositions.
• Human anatomy: Rose from #6 to #3 on hands, faces, body proportions, and dynamic poses, placing it roughly level with GPT Image 2 on its high setting.
• Text rendering: Posted a substantial gain on long passages, small text, symbols, and artistic lettering, although Artificial Analysis did not publish a rank change.
• Scene and style editing: Moved from #24 to just behind the leading model on relighting, restyling, and background changes.
• Identity-preserving editing: Advanced from #12 to #3 when retaining faces and characters through an edit.

Lighting remains its weakest text-to-image capability relative to leading models. Enhancement and restoration ranks #18 for editing, essentially unchanged from Nano Banana 2. Google’s model card also warns that some domains may reflect knowledge only through January 2025, although Gemini 3.6 Flash lists March 2026 as its overall knowledge cutoff.

Product imagery and diagrams lead

On text-to-image tasks, Nano Banana 2.1 ranks #2 for productivity and knowledge work, which covers diagrams, infographics, charts, and slides. It trails only GPT Image 2 on its high setting and rises from #8 for the previous model. The new release also ranks #3 for live-action film and #3 for retail and ecommerce, while architecture and real estate places #7.

Editing results show the largest use-case gains in retail and ecommerce, which climbed from #14 to #3, and marketing and advertising, which moved from #11 to #4. Those scores make the model competitive for product shots, on-model apparel, packaging mockups, and advertising creative. Social media and creator content, along with animation and gaming, remain its weakest editing categories.

Crowdsourced Arena results broadly match the benchmark findings. Nano Banana 2.1 placed #4 in multi-image editing with a score of 1431, #5 in text-to-image with 1328, and #6 in image editing with 1428. It ranked above Nano Banana Pro on the two main boards and about 96 points behind GPT Image 2.5 Sunburst, although the Arena scores remain preliminary.

Migration details and API limits

Google will stop serving gemini-3.1-flash-image on October 29, 2026. Applications must switch to gemini-nano-banana-2.1 before that deadline and account for the new model’s higher latency.

Thinking controls how much internal reasoning the model applies before producing an answer. Those tokens are billable, so the medium default can raise the effective cost above the advertised per-image rate.

Google is rolling out Nano Banana 2.1 across the Gemini app, AI Mode in Search, AI Studio, Flow, Stitch, Google Ads, Vertex AI, and the Gemini Enterprise Platform.

Production migration checklist

• Replace the model ID: Change gemini-3.1-flash-image to gemini-nano-banana-2.1 before October 29, 2026.
• Retest timeouts and throughput: Typical 1K generation time increases from 8.3 seconds to about 16 seconds.
• Set Thinking explicitly: Choose minimal, medium, or high according to quality and cost requirements.
• Update cost estimates: Include billable Thinking tokens alongside image-output charges.
• Run image regressions: Test critical prompts, particularly workflows involving lighting, enhancement, restoration, and architecture.