Overview
nano-banana-3-flash is Google’s gemini-3.1-flash-image-preview — the preview-tier successor to nano-banana. Same edit-mode contract (image-in + prompt → image-out), same flat $0.046 pricing, sharper output. Routed to Vertex’s global region (the regional us-central1 host 404s for this preview model — Kyma handles the routing automatically).
Tagged preview — Google may rename the underlying model id at GA. Kyma will swap the provider id transparently; your endpoint_id stays nano-banana-3-flash.
Specs
| Field | Value |
|---|---|
| Model ID | nano-banana-3-flash |
| Creator | |
| Backend | Vertex AI (gemini-3.1-flash-image-preview, global region only) |
| Best for | Bleeding-edge image edit + gen, newer quality |
| Edit mode | Yes — pass image_url to enable |
| Release stage | Preview |
| Pricing mode | Flat per image (1K tier default) |
| Default latency | ~5–10s |
| Output | Blob-hosted URL (Vercel CDN) |
Pricing
| Variant | Provider cost | Kyma list |
|---|---|---|
nano-banana-3-flash | $0.034 | $0.046 |
nano-banana. Vertex’s image size control isn’t honored today — every call returns ~1K output regardless. Will switch to per-quality tiering when Vertex ships size control.
Compared to nano-banana stable
| Strength | nano-banana-3-flash | nano-banana |
|---|---|---|
| Edit-mode quality | ★★★★★ (sharper) | ★★★★ |
| Stability | preview | stable |
| Cost | identical | identical |
| Latency | identical | identical |
Use this when
- You want the freshest Google image model and can absorb preview-tier risk (model id may change at GA).
- Edit-mode output quality matters more than stability.
Pick something else when
- You need a production SLA → use
nano-bananauntil 3-flash promotes to stable. - See
nano-bananafor the rest of the decision tree.
Example
GET /v1/jobs/{id} until succeeded.