Nano Banana 2.1 Specs and API Pricing

TL;DR

Nano Banana 2.1 reaches an Elo preference score of 1050 ± 14 compared to its predecessor's ~990 score. Pricing starts at $0.08 for 1K outputs and caps at $0.16 for 4K. The model adds 131,072 context tokens, groundings via search, and multi-turn character consistency.

What is new in Nano Banana 2.1?

Google DeepMind released Nano Banana 2.1 on October 6, 2026. Built on the Gemini 3.6 Flash lineage, it keeps the speed profile of Nano Banana 2 while updating rendering capabilities across 1K, 2K, and 4K output resolutions.

The input context window holds 131,072 tokens with a maximum output limit of 32,768 tokens. Developers can pass long prompts alongside multiple visual references in a single call.

{
  "model": "gemini-nano-banana-2.1",
  "prompt": "Create an infographic summarizing quarterly energy production.",
  "generation_config": {
    "resolution": "4K",
    "thinking_level": "medium"
  }
}

Multi-image fusion supports up to 14 reference images simultaneously. Within a single generation, you can keep up to 4 characters consistent and track up to 10 specific objects. The API also supports multi-turn character consistency across sequential image edits, alongside mask-based editing options.

Rendering extreme aspect ratios like 1:4, 4:1, 1:8, and 8:1 at 2K and 4K resolution no longer generates tiling artifacts.

How does Nano Banana 2.1 rank on Elo?

Benchmark data published by Google DeepMind rates Nano Banana 2.1 at an Overall Preference (text-to-image) score of Elo 1050 ± 14.

On the Infographic Design benchmark, the model scored 1048 ± 17. That improves on Nano Banana 2's ~961 rating through tighter layout placement and text accuracy.

Visual factuality metrics changed considerably. The model reached an Infographic Factuality score of 0.521, up from 0.179 on Nano Banana 2 and 0.265 on Nano Banana Pro.

How much does the API cost?

Pricing uses a fixed per-image rate tied to target output resolution.

Generating or editing a 1K image costs $0.08. Moving to 2K costs $0.12, and 4K outputs cost $0.16 per image.

The API exposes three thinking levels: minimal, medium (default), and high. Choosing high thinking mode adds $0.002 per image.

Call the API using model ID gemini-nano-banana-2.1. It is available across Google AI Studio, the Gemini API, the Gemini app, Google Flow, Gemini Enterprise Agent Platform, Google Ads, Google Search AI Mode, and Google Stitch.

Does it support real-time web grounding?

Nano Banana 2.1 grounds prompts using Google Web and Image Search.

{
  "model": "gemini-nano-banana-2.1",
  "prompt": "Chart the current stock price of Alphabet alongside major product milestones.",
  "grounding_config": {
    "enable_web_search": true
  }
}

Enabling search grounding costs $0.015 per request. Grounding pulls web data into the visual generation step, which supports the model's 0.521 Infographic Factuality score.

The model comes with clear functional limits:

• Audio generation is not supported
• Code execution is not supported
• Function calling is not supported
• Live API inference is not supported
• Structured outputs are not supported
• URL context is not supported

Left and right placements can cause occasional spatial localization issues. Tasks requiring true 3D reasoning or deep world knowledge remain limited.

Is upgrading to 2.1 worth it?

Pipelines rendering visual text, infographics, or consistent character sequences will see measurable improvements over Nano Banana 2.

Moving from a 0.179 to a 0.521 Infographic Factuality score reduces manual review overhead on data graphics. Passing up to 14 reference images unlocks complex asset compositions.

Stick with 1K outputs at $0.08 per image for basic draft generations.

If you need widescreen 4:1 or 8:1 layouts without seams, multi-turn editing, or web-grounded text elements, Nano Banana 2.1 justifies its price points.