Nano Banana 2.1 is Google’s October 6, 2026 update to Nano Banana 2 — the model Google lists in the API as Gemini 3.1 Flash Image. It keeps the Flash-tier speed and low cost that made Nano Banana 2 popular, and improves visual quality, prompt adherence, multi-turn character consistency and text rendering. The API model ID is gemini-nano-banana-2.1, and Google describes it as the more efficient counterpart to Nano Banana Pro (Gemini 3 Pro Image).
The pitch is cost-adjusted quality. It does not beat OpenAI’s new GPT Image 2.5 on crowd-voted rankings — Arena puts it sixth on text-to-image — but at $0.0336 per 1K image it costs about one-sixth as much per image on Artificial Analysis’s default test, and it is the best-ranked Google image model on every Arena board.
What changed from Nano Banana 2
- Better image quality and realism across 1K, 2K and 4K output (1K is the default).
- No more tiling artifacts on wide and panoramic ratios — 1:4, 4:1, 1:8 and 8:1 — at 2K and 4K.
- Sharper text rendering and infographic layout, a weak spot for most image models.
- Multi-image fusion with up to 14 reference images: character consistency for up to 4 characters and object fidelity for up to 10 objects.
- Grounding with Google Web and Image Search, so generations can reflect current, real-world information.
- Configurable thinking — minimal, medium (default) and high.
Pricing
Google bills images by output tokens. All figures below are Standard rates on the paid tier; there is no free tier for this model in the API.
| Model | 1K image | 2K image | 4K image | Text input (per 1M) |
|---|---|---|---|---|
| Nano Banana 2.1 | $0.0336 | $0.0504 | $0.113 | $1.50 |
| Nano Banana 2 (Gemini 3.1 Flash Image) | $0.067 | $0.101 | $0.151 | $0.50 |
| Nano Banana Pro (Gemini 3 Pro Image) | $0.134 | $0.134 | $0.24 | $2.00 |
Batch halves the image output rate to $15 per 1M tokens. Text and thinking output is $7.50 per 1M. Grounding includes 5,000 free search requests a month, shared across all Gemini 3 and newer models, then $14 per 1,000 requests. One watch-out: some articles quote a lower 4K price than Google’s page — the official figure is 3,780 tokens, or $0.113.
How it ranks
| Leaderboard (early October 2026) | Nano Banana 2.1 | Leader |
|---|---|---|
| Arena text-to-image (Elo) | 1328 (#6, preliminary, ~5,300 votes) | GPT Image 2.5 Sunburst 1425 |
| Arena image edit (Elo) | 1428 (#6) | GPT Image 2.5 Sunburst, 1524 |
| Arena multi-image edit (Elo) | 1431 (#4) | GPT Image 2.5 Sunburst, 1527 |
| Artificial Analysis text-to-image (Elo) | 1159 (#4) | GPT Image 2.5 Sunburst (max) 1199 |
| Artificial Analysis editing (Elo) | 1140 (tied #3 with MAI-Image-2.6) | GPT Image 2.5 Sunburst (max) 1184 |
| Artificial Analysis price per 1,000 images | $33.6 | $210.7 for GPT Image 2.5 (max) |
The Arena image-edit rows come from Atlas Cloud’s October 9 write-up of Arena’s boards; the other rows are from the leaderboards linked in the sources. Google’s own model card compares Nano Banana 2.1 only with Nano Banana 2 and Nano Banana Pro, and on those it wins — the thinking variant scores 1050 on overall text-to-image preference versus 990 for Nano Banana 2 and 935 for Pro, according to Atlas Cloud’s reading of the card. Arena agrees that it is the strongest Google image model: it sits 41 to 66 points above Nano Banana 2 across the three boards.
Specifications
- Model code:
gemini-nano-banana-2.1(stable). Last updated October 2026. - Inputs: text, image, video, PDF. Outputs: image and text.
- Limits: 131,072 input tokens, 32,768 output tokens.
- Resolutions: 1K (default), 2K, 4K.
- Supported: image generation, search grounding, thinking, Batch API.
- Not supported: function calling, structured outputs, context caching, code execution, URL context, file search, Google Maps grounding, Live API, audio generation, Flex and Priority inference.
Availability and migration
In the Gemini app, it is not a free upgrade. Press reports citing Google’s Gemini image page say Google AI Plus, Pro and Ultra subscribers get Nano Banana 2.1, while free accounts keep an earlier Nano Banana model (reports differ on whether that is Nano Banana 2 or a Lite variant). The Gemini API has no free tier for 2.1 either, so there is no official free route. The rollout was still in progress in early October, so check Google’s current help pages.
Google’s deprecations page lists gemini-nano-banana-2.1 as the recommended replacement for gemini-3.1-flash-image (Nano Banana 2), gemini-3.1-flash-image-preview and the older gemini-2.5-flash-image-preview and Imagen 4 IDs. It shows no shutdown date for Nano Banana 2 itself, so migrate on your own schedule — but retest prompts first, since output style shifts between versions.
How it compares
- vs GPT Image 2.5: GPT Image 2.5 ranks higher on every public board and handles edits with tighter control; Nano Banana 2.1 is much cheaper per image, offers search grounding and supports extreme 1:8 and 8:1 aspect ratios, where GPT Image 2.5 is limited to 1:3 to 3:1.
- vs Nano Banana 2: better quality and half the output price per image; pay more for input tokens and expect different style at the margins.
- vs Midjourney v8.1: Midjourney leads on stylized, art-directed output but has no official API; Nano Banana 2.1 is the API-native choice.
- vs FLUX.2: FLUX.2 can be self-hosted; Nano Banana 2.1 is hosted only and wins on integrated search grounding and reference fusion.
Limitations
- Early, preliminary rankings — a few thousand votes against tens of thousands for older models.
- Narrow API feature set — no function calling, structured outputs or caching, so it fits pipelines that treat it as a pure image endpoint.
- Closed and metered — no weights, no free API tier.
See the image models directory and our Best AI Image Generator ranking for alternatives.
Best for
- High-volume product, marketing and social imagery where cost per image matters most
- Conversational editing and multi-turn character consistency
- Infographics, posters and diagrams that need legible rendered text
- Search-grounded visuals tied to current events, real places or real products
Pros & cons
Strengths
- Output pricing is about half of Nano Banana 2 per image ($0.0336 vs $0.067 at 1K, $0.113 vs $0.151 at 4K) and roughly one-sixth of GPT Image 2.5 at max quality on Artificial Analysis's default test
- Google's most capable Flash-tier image model — the highest-ranked Google entry on Arena's text-to-image, image-edit and multi-image-edit boards, ahead of Nano Banana 2 and Nano Banana Pro
- Fuses up to 14 reference images (up to 4 characters and 10 objects) and grounds generation in Google Web and Image Search
- Three thinking levels (minimal, medium by default, high), plus a fix for tiling artifacts on very wide and tall aspect ratios at 2K and 4K
Limitations
- Trails OpenAI's GPT Image 2.5 pair and GPT Image 2 on Arena's text-to-image board (1328 vs 1425 / 1398 / 1383), and its entries are still preliminary with only about 5,300 votes
- Text input costs 3x Nano Banana 2 ($1.50 vs $0.50 per 1M tokens), which narrows savings on prompt-heavy and many-reference workloads
- No function calling, structured outputs, context caching, Flex or Priority inference, and no free tier in the paid API
- Closed model — Google's own model card compares it only with other Google image models
Frequently asked questions
What is Nano Banana 2.1?
Nano Banana 2.1 (API ID gemini-nano-banana-2.1) is Google's image generation and conversational editing model, released October 6, 2026. It updates Nano Banana 2, which Google lists as Gemini 3.1 Flash Image, keeping Flash-level speed and cost while improving visual quality, prompt adherence, multi-turn character consistency and text rendering.
How much does Nano Banana 2.1 cost?
On the Gemini API paid tier, images cost $30 per 1M output tokens ($15 with Batch), which Google converts to $0.0336 per 1K image, $0.0504 per 2K image and $0.113 per 4K image. Input is $1.50 per 1M tokens for text, image and video. Some third-party articles quote a lower 4K price — Google's pricing page lists $0.113.
Is Nano Banana 2.1 better than GPT Image 2.5?
Not on crowd-voted rankings. Arena places Nano Banana 2.1 sixth on text-to-image at 1328, behind GPT Image 2.5 Sunburst (1425), Flare (1398) and GPT Image 2 (1383), and Artificial Analysis shows the same order. It is far cheaper per image, though, and it is the stronger Google model — it leads Nano Banana 2 and Nano Banana Pro on Arena.
What are the differences between Nano Banana 2.1 and Nano Banana 2?
Nano Banana 2.1 improves visual quality at 1K, 2K and 4K, fixes tiling artifacts on 1:4, 4:1, 1:8 and 8:1 aspect ratios, sharpens text and infographic layout, and adds configurable thinking levels. It also halves the per-token image output price ($30 vs $60 per 1M) while tripling text input price ($1.50 vs $0.50).
Is Nano Banana 2 being shut down?
Not yet. Google's deprecations page lists gemini-nano-banana-2.1 as the recommended replacement for gemini-3.1-flash-image but shows no shutdown date for either model. Some outlets have reported a date — check Google's page before planning a migration.
How many reference images can Nano Banana 2.1 use?
Up to 14 in a single request, with character consistency for up to 4 characters and object fidelity for up to 10 objects.
Sources
- Google AI for Developers — Gemini Nano Banana 2.1 model page
- Google AI for Developers — Gemini Developer API pricing
- Google AI for Developers — Deprecations
- Arena — Text-to-Image leaderboard
- Artificial Analysis — AA-Image-T2I v2.0 leaderboard
- Artificial Analysis — AA-Image-Editing v2.0 leaderboard
- Atlas Cloud — Nano Banana 2.1 vs GPT Image 2.5
- MadRobot — Nano Banana 2.1: Google's cheaper image model, prices and access