Nano Banana 2.1 Model Card - Multi-Reference Editing: leaderboard
Multi-Reference Editing. Google DeepMind's October 6, 2026 Nano Banana 2.1 model card compares four image-generation configurations. Nano Banana 2.1 Thinking and No Thinking remain separate; No Thinking is not assumed to mean an API minimal-thinking setting. The thinking budget is unspecified. Gemini 3.1 Flash Image is the Thinking comparator; Gemini 3 Pro Image has no stated thinking setting. The comparator labels do not identify stable versus preview API snapshots, so their source-qualified identities remain separate from exact endpoints. The evaluation uses mixed public and internal capability sets; exact scored prompts, reference images, sample sizes, and AutoRater version are not supplied. These source-specific comparisons are not fused with another image leaderboard. The native score is Elo from side-by-side human evaluation. The displayed plus/minus uncertainty is not added to the score, and Elo has no fixed theoretical endpoints. Reported metric: Human evaluation Elo.
Metric: Human evaluation Elo. Source: deepmind.google. Saturation forecast: Rough model projection: around 2026. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gemini-nano-banana-2.1 (Thinking) | 1066 | |
| 2 | gemini-nano-banana-2.1 (Non-Thinking) | 1041 | |
| 3 | Gemini 3 Pro Image (Nano Banana 2.1 Model Card) | 989 | |
| 4 | Gemini 3.1 Flash Image (Thinking) (Nano Banana 2.1 Model Card) | 988 |
Interactive version: theaggregate.ai/benchmark?slug=nano-banana-2-1-model-card-multi-reference-editing · How It Works · Data refreshed daily, snapshot 2026-10-25.