Google Moves Nano Banana 2 and Nano Banana Pro to Full Production
Google's Nano Banana 2 and Nano Banana Pro image generation models hit GA with production SLAs, new video input, and ~$0.07/image pricing
Takeaways
- GA launch: Nano Banana 2 (
gemini-3.1-flash-image) and Nano Banana Pro (gemini-3-pro-image) are now production-ready with enterprise SLAs via Google Cloud. - Deprecation deadline: Preview endpoints (
*-previewsuffix) retire June 25, 2026 — update your API calls now. - New capability: Nano Banana 2 now accepts video files (including YouTube URLs) as input to generate thumbnails, posters, and infographics.
- Pricing: Nano Banana 2 costs ~$0.067 per 2K image (50% cheaper than Pro); batch mode halves that again to ~$0.034.
- Benchmarks: Nano Banana 2 ranked #1 in the Artificial Analysis Text-to-Image Arena at roughly half the price of GPT Image 1.5.
- Access: Available via Gemini API (no SLA) or Vertex AI / Gemini Enterprise Agent Platform (full enterprise SLA).
Google has officially moved its two flagship image generation models, Nano Banana 2 and Nano Banana Pro, to General Availability. That means stable model IDs, production-grade SLAs, and no more -preview suffix in your API calls. For teams that have been holding off on committing these models to production pipelines, the wait is over.
The two models map to distinct positions in Google's image generation lineup: Nano Banana 2 is gemini-3.1-flash-image, built for speed and high-volume workloads, while Nano Banana Pro is gemini-3-pro-image, designed for professional-grade precision and complex creative tasks.
Google's product strategy here is a deliberate dual-model approach: one focused on extreme quality (Pro) and the other on extreme efficiency (Flash).
What GA actually means for your code
On May 28, Google DeepMind officially promoted both gemini-3.1-flash-image-preview and gemini-3-pro-image-preview to General Availability. Developers no longer need to use temporary IDs with the -preview suffix in production environments and can now call the stable versions directly.
If you're still on the preview endpoints,
the preview versions will retain a roughly four-week buffer period after GA, with official retirement on June 25. Any code in production still using the *-preview suffix must be updated before that date to avoid 404 errors.
This isn't just a naming change. Preview models are generally not recommended for production environments due to lack of SLA guarantees and potential API behavior changes. Now that both models come with production-grade SLAs, application developers can confidently integrate them into their core business workflows. Before GA, developers were regularly hitting 503 overload errors during peak times as Google allocated limited compute to preview endpoints. That structural problem goes away with GA.
The models, side by side
Here is how the two models compare across the dimensions that matter most for production decisions:
| Model | API ID | Speed | Best for | Price (2K image) |
|---|---|---|---|---|
| Nano Banana 2 | gemini-3.1-flash-image | <2s | High-volume, fast iteration | ~$0.067 |
| Nano Banana Pro | gemini-3-pro-image | 2-5s | Text rendering, complex scenes | ~$0.134 |
Image generation with Nano Banana 2 costs 50% less than Nano Banana Pro, while text input tokens are 75% cheaper. Despite the reduced cost, the model delivers what Google describes as "Pro-level visual quality at Flash speed,