Google released Nano Banana 2.1 on October 6, 2026: an update to its Nano Banana 2 image model that halves the API price of an image, to $0.0336 at 1K, and entered Arena’s text-to-image leaderboard at #5. Google’s launch post on X says the model “outperforms our previous models across the board, with notable leaps in visual design, mask-based editing, and subject consistency”. It is generally available in the Gemini API as gemini-nano-banana-2.1, and Nano Banana 2 shuts down on October 29. This page reads Google’s API changelog, model page, pricing page and Arena’s text-to-image and image edit leaderboards, as of October 6, 2026.
On this page · 6 sectionsOpen
Google released Nano Banana 2.1 on October 6, 2026, an update to its Nano Banana 2 image model, generally available in the Gemini API as gemini-nano-banana-2.1 and rolling out to the Gemini app, AI Mode in Search, AI Studio, Flow, Stitch and Google Ads.
Image output now costs $30 per million tokens instead of $60: $0.0336 for a 1K image, $0.0504 at 2K and $0.0756 at 4K on the Gemini API, half of Nano Banana 2 and a quarter of Nano Banana Pro at 1K.
Arena placed it #4 in Multi-Image Edit (1431), #5 in Text-to-Image (1328) and #6 in Image Edit (1428), ahead of Nano Banana Pro on both main boards and about 96 points behind GPT Image 2.5 Sunburst; the scores are preliminary.
It fuses up to 14 reference images (up to 4 characters and 10 objects), fixes tiling artifacts on 1:8 and 8:1 panoramas at 2K and 4K, and adds minimal, medium and high thinking levels.
Nano Banana 2 (gemini-3.1-flash-image) is deprecated and shuts down on October 29, 2026, so builders have 23 days to switch the model ID.
Input tripled, from $0.50 to $1.50 per million tokens, and Google Cloud’s spec counts 3,780 tokens for a 4K image against 2,520 on the Gemini API, so check both before budgeting.
§ 01What Google shipped
The changelog calls Nano Banana 2.1 “the latest high-efficiency image generation and conversational editing model” and says it “maintains Flash-level speed and cost efficiency”. Google’s thread says it “starts rolling out today” across the Gemini app, AI Mode in Search, Google AI Studio, Flow, Stitch, Google Ads and the Gemini Enterprise Platform. No post on blog.google or the DeepMind site had appeared by the evening of October 6; the API docs and the X thread are the record.
| Item | Nano Banana 2.1 |
|---|---|
| Release | October 6, 2026, generally available |
| API model ID | gemini-nano-banana-2.1 |
| Replaces | Nano Banana 2 (gemini-3.1-flash-image), shut down October 29, 2026 |
| Inputs | Text, image, video, PDF |
| Outputs | Image and text |
| Resolutions | 1K (default), 2K, 4K |
| Aspect ratios | 15, from 1:1 to 1:8 and 8:1 |
| Reference images | Up to 14: up to 4 characters and up to 10 objects |
| Thinking levels | Minimal, medium (default), high |
| Grounding | Google Web and Image Search |
| Token limits | 131,072 input, 32,768 output |
The model page lists five concrete changes over Nano Banana 2: better visual quality and realism at all three resolutions, a fix for tiling artifacts on the wide and tall ratios (1:4, 4:1, 1:8, 8:1) at 2K and 4K, better text rendering and infographic layout, multi-image fusion from up to 14 references, and configurable thinking. Mask-based editing is named in Google’s X post but has no parameter or benchmark on the API pages we read, so treat it as a quality claim for now. The model page also lists what it does not do: no function calling, no structured outputs, no context caching, no Live API.
§ 02What it costs
Image output is billed per token, and Google prices a 1K image at 1,120 output tokens, a 2K image at 1,680 and a 4K image at 2,520. Nano Banana 2.1 charges $30 per million image output tokens; Nano Banana 2 charged $60. That is the halving.
| Model | Input per 1M tokens | Image output per 1M tokens | 1K image | 2K image | 4K image |
|---|---|---|---|---|---|
| Nano Banana 2.1 | $1.50 | $30 | $0.0336 | $0.0504 | $0.0756 |
| Nano Banana 2 | $0.50 | $60 | $0.067 | $0.101 | $0.151 |
| Nano Banana Pro | $2.00 | $120 | $0.134 | $0.134 | $0.24 |
Three details change the math:
- Input tripled. Text, image and video input went from $0.50 to $1.50 per million tokens. A text prompt of 200 tokens costs $0.0003, so for a plain generation the output is the bill. An edit that feeds in many reference images pays three times as much for that input.
- Batch is half again. The Batch API prices image output at $15 per million: $0.0168 at 1K, $0.0252 at 2K, $0.0378 at 4K.
- 4K on Google Cloud reads differently. The Google Cloud model spec counts 3,780 output tokens for a 4K image, against 2,520 on the Gemini API pricing page. At $30 per million that is $0.1134 against $0.0756. Check your Cloud invoice before you budget 4K at the API figure.
For scale against OpenAI: we measured GPT Image 2.5 at $0.013 per 1024 by 1024 image on medium quality and $0.053 on high (our September 8 test). Nano Banana 2.1’s 1K price sits between the two.
§ 03Where it ranks on Arena
Arena’s post places Nano Banana 2.1 at #4 in Multi-Image Edit with 1431 points, #5 in Text-to-Image with 1328 and #6 in Image Edit with 1428. Arena calls it “a substantial improvement from Nano Banana 2” and gives the gains as 80 points in text-to-image, 67 in multi-image edit and 38 in image edit. All three scores are marked preliminary: the text-to-image entry had 5,312 votes, against 17,103 for the leader.
| Model | Text-to-image rank | Text-to-image score | Image edit rank | Image edit score |
|---|---|---|---|---|
| GPT Image 2.5 Sunburst | 1 | 1425 | 1 | 1524 |
| GPT Image 2.5 Flare | 2 | 1398 | 2 | 1481 |
| GPT Image 2 (medium) | 3 | 1383 | 3 | 1462 |
| MAI Image 2.6 | 4 | 1333 | 5 | 1428 |
| Nano Banana 2.1 | 5 | 1328 | 6 | 1428 |
| Grok Imagine Image 2.0 (canvas) | 6 | 1321 | 4 | 1439 |
| Nano Banana 2 (web search) | 11 | 1261 | 14 | 1387 |
| Nano Banana Pro (2K) | 16 | 1248 | 12 | 1390 |
Two readings from the table. First, Google’s cheap model now outscores its expensive one: Nano Banana 2.1 beats Nano Banana Pro by 80 points in text-to-image and 38 in image edit, at a quarter of Pro’s 1K price. Second, it trails GPT Image 2.5 Sunburst by 97 points in text-to-image and 96 in image edit. Arena’s stated gains compare against its own Nano Banana 2 entries; the web-search entry shown here gives 67 and 41 by subtraction.
§ 04What changes if you build on it
- Migrate before October 29. The changelog marks
gemini-3.1-flash-imagedeprecated with shutdown on October 29, 2026. The swap is the model ID,gemini-nano-banana-2.1, then a check of your input costs. - The smallest size is now 1K. Nano Banana 2 listed a 0.5K image at $0.045. Nano Banana 2.1 lists 1K, 2K and 4K only, and its 1K image costs less than the old 0.5K one.
- Panoramas are usable at high resolution. The tiling fix at 2K and 4K on the 1:8 and 8:1 ratios matters for banners and strips, the formats where earlier models repeated patterns.
- Search grounding costs extra. 5,000 grounded requests a month are free across Gemini 3.x models, then $14 per 1,000.
§ 05What we are watching
- Arena’s preliminary scores. With about 5,000 text-to-image votes, the 1328 can still move by several points either way.
- A Google benchmark post. Google has published no scores of its own for Nano Banana 2.1 yet; a blog.google or DeepMind post would carry them.
- The 4K token count. Whether the Gemini API page (2,520 tokens) or the Cloud spec (3,780) changes to match the other.
- A Pro successor. Nano Banana Pro now trails the Flash-tier model on both boards.
§ 06Sources
- Google on X, launch post and rollout thread, October 6, 2026: x.com/google
- Gemini API changelog, October 6, 2026: ai.google.dev
- Gemini Nano Banana 2.1 model page: ai.google.dev
- Gemini API pricing: ai.google.dev
- Google Cloud model spec, Nano Banana 2.1: docs.cloud.google.com
- Arena on X, October 6, 2026: x.com/arena
- Arena leaderboards: text-to-image, image edit
Q1What is Nano Banana 2.1?
Nano Banana 2.1 is Google’s image generation and conversational editing model, released and generally available on October 6, 2026. It updates Nano Banana 2 (gemini-3.1-flash-image) with better visual quality, prompt adherence, character consistency and text rendering. Its API model ID is gemini-nano-banana-2.1.
Q2How much does Nano Banana 2.1 cost?
On the Gemini API standard tier, image output costs $30 per million tokens: $0.0336 for a 1K image, $0.0504 at 2K and $0.0756 at 4K. Input costs $1.50 per million tokens. The Batch API halves image output to $0.0168, $0.0252 and $0.0378.
Q3Is Nano Banana 2.1 better than GPT Image 2.5?
Not on Arena. On October 6, 2026 GPT Image 2.5 Sunburst led text-to-image at 1425 and image edit at 1524; Nano Banana 2.1 scored 1328 and 1428, preliminary. Nano Banana 2.1’s 1K price of $0.0336 sits between GPT Image 2.5’s measured medium ($0.013) and high ($0.053) costs.
Q4When does Nano Banana 2 shut down?
Google’s changelog says gemini-3.1-flash-image is deprecated and shuts down on October 29, 2026. Google recommends migrating to gemini-nano-banana-2.1, which leaves 23 days from the October 6 release.
Q5Is Nano Banana 2.1 better than Nano Banana Pro?
On Arena’s October 6 boards, yes: 1328 against 1248 in text-to-image and 1428 against 1390 in image edit, at $0.0336 per 1K image against Pro’s $0.134. Pro remains Google’s higher-priced model; Google has not announced a successor.