Gemini 2.5 Flash Image Preview (nano-banana)

GoogleProprietary

Gemini 2.5 Flash Image Preview (nano-banana) is an AI image model from Google. On LMArena’s Text-to-image leaderboard it ranks 34 of 82 with 1150±2 from 890,539 votes. On LMArena’s Image editing leaderboard it ranks 32 of 60 with 1291±2 from 11,430,191 votes. License: Proprietary.

Key figures

LMArena · Text-to-image1150±2#34 / 82 · Votes 890,539
LMArena · Image editing1291±2#32 / 60 · Votes 11,430,191
Developer
Google
License
Proprietary

Not yet available: Released (per AA), AA Elo

Who it is for (from the data)

  • Solid quality with room to choose: #34 of 82 on Text-to-image, in the top half.
  • Generating and editing images: it is ranked on both boards.
  • Use via Google's service or API: LMArena lists it as proprietary.
  • Existing Google users: this vendor has 10 models on Text-to-image.

Neighbours on Text-to-image

#ModelScoreVotesAA EloLicenseΔ
31Reve V1.5 Reve115539,700–Proprietary+5
32FLUX 2 Pro Black Forest Labs1153212,359–Proprietary+3
33Cosmos3-Super-Text2Image NVIDIA115218,485–OpenMDW-1.1+2
34Gemini 2.5 Flash Image Preview (Nano Banana) Google1150890,539–Proprietary
35Hunyuan Image 3.0 Tencent1150180,518989tencent-hunyuan-community+0
36Seedream 4.5 ByteDance1149300,1071026Proprietary-1
37Imagen Ultra 4.0 Generate 001 Google1148388,736–Proprietary-2

Alternatives · More from Google

#ModelScoreVotesAA EloLicense
6Gemini Nano Banana 2.1 Google13285,3101176Proprietary
11Gemini 3.1 Flash Image (Nano Banana 2) [Web Search] Google126260,515–Proprietary
15Gemini 3.1 Flash Lite Image (Nano Banana 2 Lite) Google125017,804–Proprietary
16Gemini 3 Pro Image 2K (Nano Banana Pro) Google1248175,734–Proprietary
18Gemini 3 Pro Image Preview (Nano Banana Pro) Google123486,464–Proprietary

Alternatives · Top open-weight models

#ModelScoreVotesAA EloLicense
19Qwen Image 2.1 Alibaba122311,4411042qwen-research
20Ideogram 4.0 Quality Ideogram120560,591–Ideogram Open Model
27Cosmos3-Super-Text2Image (Agentic) NVIDIA116510,805–OpenMDW-1.1
33Cosmos3-Super-Text2Image NVIDIA115218,485–OpenMDW-1.1
35Hunyuan Image 3.0 Tencent1150180,518989tencent-hunyuan-community

How to read the scores

  • LMArena score: people blind-pick the better of two anonymous images for the same prompt, and the votes are fitted into an Elo-style rating; "±" is the 95% confidence interval.
  • Text-to-image and image edit are separate boards, so scores are not comparable across them; AA Elo comes from Artificial Analysis on a different scale.
  • If the gap between two models is smaller than their combined intervals, the order is not statistically reliable; treat them as the same tier. More votes mean narrower intervals.

FAQ

Where does Gemini 2.5 Flash Image Preview (nano-banana) rank among AI image models?

Text-to-image rank 34 (1150); Image editing rank 32 (1291) (LMArena, data as of Oct 26, 2026).

Is Gemini 2.5 Flash Image Preview (nano-banana) open-weight?

No. LMArena lists it as Proprietary; it is available only through Google’s products or API.

Can Gemini 2.5 Flash Image Preview (nano-banana) edit images?

Yes. Gemini 2.5 Flash Image Preview (nano-banana) is #32 of 60 on the LMArena image-edit board (1291, 11,430,191 votes).

Who makes Gemini 2.5 Flash Image Preview (nano-banana)?

According to LMArena, Gemini 2.5 Flash Image Preview (nano-banana) is made by Google.

How reliable is Gemini 2.5 Flash Image Preview (nano-banana)'s score?

The scores are based on 890,539 votes on Text-to-image (95% CI ±2); 11,430,191 votes on Image editing (95% CI ±2). More votes and narrower intervals mean a more stable rank.

How far is Gemini 2.5 Flash Image Preview (nano-banana) from #1 on Text-to-image?

#1 on Text-to-image is GPT Image 2.5 Sunburst (1425); Gemini 2.5 Flash Image Preview (nano-banana) scores 1150, 275 points behind.

Sources

LMArena Text-to-image data published on Oct 7, 2026.

LMArena Image editing data published on Oct 20, 2026.

17NAS syncs the upstream leaderboards daily; this page's data was synced on Oct 26, 2026.

Price, resolution and speed have no reliable public data yet and are omitted; all values come from the snapshot, blanks show –.