Skip to main content
  • Published: July 22, 2026
  • Last verified: July 22, 2026 (at publication)
  • Status: Archived; this page records the state on July 22, 2026
On July 22, 2026, gpt-image-2-vip supported 4K sizes and the quality parameter again, and the Nano Banana Pro, Nano Banana 2, and Nano Banana 2 Lite routes were stable for professional final assets, general high-throughput generation, and lower-cost bulk work. For newer image model announcements, see the GPT Image 2.5 launch and the Grok Imagine Image 2.0 launch. Current prices are in the model and pricing catalog.

​
What changed?

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

​
Who this affected

  • Users of gpt-image-2-vip in the default group who needed 4K output or the quality parameter.
  • Teams choosing among Nano Banana Pro, Nano Banana 2, and Nano Banana 2 Lite for high-concurrency production.
Other image models, text models, and video APIs were not affected by this announcement.

​
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: a per-call token in the default group
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. The Nano Banana API overview compares the current models.

​
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” meant that the route could serve production traffic. It did not mean unlimited concurrency or zero errors. Effective throughput depends on token group, upstream capacity, input size, and content-safety policies.

​
Frequently asked questions

​
Did gpt-image-2-vip support 4K and quality at publication?

Yes. On July 22, 2026, gpt-image-2-vip supported common 4K output and the quality parameter again. For current behavior, see the GPT Image 2 API guide, and 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.

​
How do Nano Banana Pro, Nano Banana 2, and Lite differ?

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.

​
Are the prices on this page still current?

Not necessarily. The table records prices at publication; the model and pricing catalog lists current rates. 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 keep 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