Skip to main content
  • Published: July 22, 2026
  • Status verified: July 22, 2026 (at publication)
  • Announcement status: Restored and production-ready at publication; current access and billing are shown in the console and call logs
This announcement records that gpt-image-2-vip supported 4K sizes and the quality parameter again at publication. Nano Banana Pro, Nano Banana 2, and Nano Banana 2 Lite routes were also stable for professional final assets, general high-throughput generation, and lower-cost bulk work.

​
What changed?

These prices record the LaoZhang API per-call rates at publication. Check the console model and pricing page and call logs for current prices, token groups, and actual charges.

​
How do I integrate GPT-Image-2-VIP?

  • Image generation: POST /v1/images/generations
  • Image editing: POST /v1/images/edits
  • Model ID: gpt-image-2-vip
  • Sizes: common 1K / 2K / 4K outputs; see the integration guide for exact parameters
  • Quality values: low, medium, high
  • Token requirement: Default Group per-request billing token
If a request does not return the expected size or quality, first check the effective model, token group, request body, and error in call logs. Then compare the payload with the GPT Image 2 API guide.

​
Which Nano Banana model should I choose?

  • Test gemini-3-pro-image first for professional design, text layout, advanced editing, or 4K final assets.
  • Test gemini-3.1-flash-image first for a balance of quality, speed, and price in general production.
  • Test gemini-3.1-flash-lite-image first for 1K drafts, bulk assets, or low-latency high concurrency.
Some console groups may still show gemini-3-pro-image-preview or gemini-3.1-flash-image-preview as compatibility aliases. These are not preferred long-term IDs; new projects should use the stable model IDs currently listed in the console.

​
Pre-production checklist for high concurrency

  1. Test generation and editing with the exact production token.
  2. Cover the sizes, aspect ratios, quality values, and source-image sizes used by the application.
  3. Increase concurrency in stages and record 429s, 5xx errors, average latency, and successful-image rate.
  4. Add bounded retries, fallback behavior, and human review for capacity fluctuations and content-safety rejections.
  5. Use call logs to verify actual charges for every model and parameter combination.
“Stable for high concurrency” means the current route can serve production traffic. It does not mean unlimited concurrency or zero errors. Effective throughput still depends on token group, upstream capacity, input size, and content-safety policies.

​
Frequently asked questions

​
Does GPT-Image-2-VIP support 4K and quality now?

Yes. This announcement confirms that gpt-image-2-vip again supports common 4K output and the quality parameter. Run a small-volume request with your own token and production payload before scaling traffic.

​
Which endpoints handle GPT-Image-2-VIP generation and editing?

Use POST /v1/images/generations for generation and POST /v1/images/edits for editing. Both endpoints use the model ID gpt-image-2-vip.

​
What is the difference between Nano Banana Pro, Nano Banana 2, and Lite?

Pro targets advanced design and high-quality final assets. Nano Banana 2 targets general high-throughput production. Lite targets 1K, low-latency, cost-sensitive bulk work. Compare quality, latency, and actual charges with your own prompts and source images.

​
Will the announced prices remain unchanged?

Not necessarily. The table records prices at publication; the console defines current rates and billing. Google’s upstream pricing and LaoZhang API per-call pricing are separate billing contracts.

​
Can I still use preview model IDs?

Some token groups may temporarily retain compatibility aliases. New projects should prefer gemini-3-pro-image, gemini-3.1-flash-image, and gemini-3.1-flash-lite-image, subject to the model list currently shown in the console.

​
Can high-concurrency requests still receive 429 or 5xx errors?

Yes. A restored, stable route does not remove token limits, upstream capacity events, content-safety rejections, or transient network errors. Use staged load tests, bounded retries, fallback behavior, and monitoring instead of moving all traffic at once.

​
Sources and integration guides