Nano Banana 2.1: Google Cuts Price in Half - Launch Day Test

👁 121 reads

Google released Nano Banana 2.1 on October 6, 2026, a new version of its image generation and editing model. In the API it’s half the price of Nano Banana 2, adds 8:1 panoramas, and improves mask editing. We tested it immediately on the same tasks as previous models - results below, and full model coverage in our Nano Banana guide.

The launch was quiet. Instead of a conference, there was an entry in the Gemini API changelog and a brief announcement from the @NanoBanana account on X, published on the evening of October 6. Google states that the new version “surpasses previous models in everything” and lists four improvements: visual design, mask editing, character and object consistency, and more natural photos. The model rolled out to the Gemini app, Google AI Studio, and “other” locations.

Key Features of Nano Banana 2.1

• API name: gemini-nano-banana-2.1. Google describes it as a performance-focused update to Nano Banana 2 (Gemini 3.1 Flash Image).
• Price: $0.0336 per 1K image, $0.0504 for 2K, $0.113 for 4K. Nano Banana 2 cost $0.067 / $0.101 / $0.151.
• New aspect ratios: 1:4, 4:1, 1:8, and 8:1 - banners, strips, and panoramas without stitching.
• References: up to 14 images in one prompt, consistency for up to 4 characters and 10 objects.
• Thinking modes: three levels - minimal, medium, and high.
• Input: text, images, video, and PDF; Google search as a source of facts for graphics.
• Access: Gemini app, AI Studio, AI mode in search, Flow, Stitch, Google Ads, and from day one also OpenRouter and fal.ai.
• Successor: gemini-3.1-flash-image (Nano Banana 2) is marked as deprecated.

Nano Banana 2.1 Pricing

The reduction applies to output images ($30 instead of $60 per million tokens). Text input increased from $0.50 to $1.50 per million tokens, and “thinking” from $3 to $7.50 - for typical prompts that’s pennies, but with a dozen reference photos it’s worth calculating. In our calls through OpenRouter, a 1K image cost $0.035-0.048, and 4K cost $0.117. In the Gemini app, the model is also available in the free plan, with a daily limit.

Arena Leaderboard Rankings

In the Arena text-to-image leaderboard from October 6, Nano Banana 2.1 entered at 5th place with a score of 1328 ± 9 (5312 votes, preliminary result). Higher are three OpenAI models and one from Microsoft:

1. GPT Image 2.5 Sunburst - 1425
2. GPT Image 2.5 Flare - 1398
3. GPT Image 2 (medium) - 1383
4. MAI Image 2.6 - 1333
5. Nano Banana 2.1 - 1328

In comments under Google’s announcement, comparisons with GPT Image 2.5 appeared most frequently. We ran those comparisons on the same prompts in a separate article: Nano Banana 2.1 vs GPT Image 2.5.

Our Launch Day Test Results

On launch day we ran the model through four standard tests from our image generator arena and two tests of new features. Summary:

• Text rendering: 13/13 across four graphics (we tested in Polish) - perfect, like Nano Banana 2 and GPT Image 2.5.
• Objects in specified locations: average IoU 0.44, 7 of 16 objects correctly placed. GPT Image 2.5: 0.54, FLUX 3: 0.89.
• Sequential edits: prompts executed, but after 8 steps consistency of untouched areas dropped to SSIM 0.41 (Nano Banana 2: 0.61). Image becomes progressively pinker with each round.
• 4K detail: file 5504 × 3072 px, actual detail at approximately 4.2 Mpx - same as its predecessor.
• Mask editing: 9.0/10 versus 7.3/10 for Nano Banana 2. This is a real improvement.
• Character consistency: 9.25/10 across four scenes, same as Nano Banana 2, but at 56% of the price.

What This Means

Nano Banana 2.1 is primarily a cheaper and slightly more capable Nano Banana 2 for fragment editing, not a new quality tier. In our tests it didn’t surpass its predecessor in text rendering, composition, or 4K detail, but was noticeably better at mask editing and equal in character consistency - at half the price. The weakest point is long series of corrections on one image: here the predecessor and GPT Image 2.5 maintained colors better.

For typical Gemini app users, the change is invisible except for prettier results when correcting fragments. For businesses and app developers, it’s primarily a bill cut in half and quick migration, since the old model is being deprecated. The shutdown date cited by some sites (October 29) didn’t appear in the documentation we reviewed - we treat it as unconfirmed.

What We Still Don’t Know

• What base model 2.1 is built on - sites mention Gemini 3.6 Flash or Gemini 3.6 Image, but Google hasn’t stated this explicitly.
• We didn’t directly verify the image editing leaderboard position; industry sites report 6th place with 1428 points.
• Google as usual doesn’t specify free Gemini plan limits with numbers.

We collected ready-to-use prompts with results in the gallery Nano Banana 2.1: 15 Prompts.

Sources verified October 7, 2026: Gemini API changelog, Gemini API pricing, gemini-nano-banana-2.1 model card, @NanoBanana post from October 6, Arena text-to-image leaderboard, OpenRouter and fal.ai model lists, our tests via OpenRouter API.

// read moreNano Banana 2.1 Benchmark: Data Charts and 4-Panel Comic StripsBenchmarksNano Banana 2.1 vs GPT Image 2.5: Same Prompt Test ResultsBenchmarksNano Banana 2.1: Price, Tests and How to Use It in Gemini for FreeAI Tools

Kinetyra