Nano Banana 3 Flash

Overview

Nano Banana 3 Flash is Google’s Gemini 3.1 image generation model. It accepts text prompts and reference images, then outputs generated or modified images. The model is optimized for speed and sits in the cheap cost tier while delivering frontier-tier visual quality. On Kyma, the model runs through an OpenAI-compatible endpoint with automatic failover across serving paths. Prompt caching is enabled, so repeated prompt prefixes are billed at this model’s cached input rate. Every response returns the exact cost in the usage.cost field and identifies the executed model via the X-Kyma-Model header. As of 14 August 2026 this model serves on Google’s generally available Gemini 3.1 image endpoint rather than the preview one it launched on. The price did not change with the move. It does not support structured JSON outputs or reasoning traces, and it is strictly an image-to-image or text-to-image model. Google plans to retire the older nano-banana model in October 2026, making this the direct migration path.

Specs

Field	Value
Model ID	`nano-banana-3-flash`
Best for	In-context image edit + gen at the newest Gemini image quality
Context window	0
Input modalities	Text, Image
Output modalities	Image
Tool calling	Yes
Structured outputs	Yes
Prompt caching	Yes
Speed	Fast
Cost band	Cheap
Release stage	Stable

Pricing

	Per 1M tokens
Input	$0.20
Output	$0.60

Use this when

Generate marketing assets — Produce high-resolution product and campaign visuals from text prompts at low cost.
Edit reference images — Modify existing photos using text instructions while preserving layout and composition.
Prototype UI mockups — Quickly render interface concepts and layout variations for early-stage design reviews.
Batch image synthesis — Run large-scale generation jobs where speed and cheap-tier pricing matter most.

Not ideal for

Do not use this model if you require structured data extraction or multi-step reasoning alongside image generation.

Pick something else when

You need vector graphic generation: use recraft-v4-vector.
You need photorealistic portrait editing: use flux-2-pro.
You need the highest possible resolution: use imagen-4-ultra.

Example

from openai import OpenAI

client = OpenAI(base_url="https://kymaapi.com/v1", api_key="ky-...")

response = client.chat.completions.create(
    model="nano-banana-3-flash",
    messages=[{"role": "user", "content": "..."}],
)

Agent query example

Ask the API which models fit, instead of hardcoding an id:

curl "https://kymaapi.com/v1/models?recommended_for=generate-images&tools=true&quality_tier=frontier-open"

FAQ

Does this model accept image inputs for editing? Yes, it supports both text prompts and image inputs for generation and modification tasks. How does prompt caching affect billing on Kyma? Repeated prompt prefixes are billed at this model’s cached input rate, reducing costs for batch jobs with shared instructions. Is this a preview release? No. It launched on Google’s preview endpoint and moved to the generally available one on 14 August 2026. The model ID you call on Kyma, nano-banana-3-flash, did not change, and neither did the price.

​Overview

​Specs

​Pricing

​Use this when

​Not ideal for

​Pick something else when

​Example

​Agent query example

​FAQ