> ## Documentation Index
> Fetch the complete documentation index at: https://docs.laozhang.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# GPT Image -vip is unstable: switch routes

> The three GPT Image -vip models are short on upstream capacity. Switch to the web route at the same price, official forwarding, or Nano Banana.

* **Published**: October 23, 2026
* **Last checked**: October 23, 2026
* **Status**: Monitoring. The three `-vip` models are still listed, but many requests fail and a quick recovery is unlikely

The three per-call GPT Image VIP models are badly short on upstream capacity, and many requests fail. The quickest fix is the web route at the same price: change `model` and nothing else. For a fixed Flare or Sunburst model or native high resolution, use official forwarding, or move to the Nano Banana models.

## Your options

| Option | What changes | Price | Best for |
| - | - | - | - |
| 1. Web route (recommended) | `model` only | \$0.03 per call, same as `-vip` | Most generation and editing jobs |
| 2. Official forwarding | A token in an official-forward group; same API | Per token | Fixed Flare or Sunburst, native high resolution, official parameters |
| 3. Nano Banana | Model and request format | \$0.045–\$0.09 per call | Jobs that can accept a different model's style |

Current prices are on [Models and pricing](/en/models.md).

### Option 1: the web route (recommended)

Keep your token and endpoint, and swap the model ID:

| Current model | Switch to |
| - | - |
| `gpt-image-2-vip` | `gpt-image-2-web` |
| `gpt-image-2.5-flare-vip` | `gpt-image-2.5-web` |
| `gpt-image-2.5-sunburst-vip` | `gpt-image-2.5-web` |

How the web route differs from `-vip`:

* It supports generation and editing on the same endpoints as `-vip`, with common `size` values such as 1024x1024, 1536x1024, and 1024x1536.
* It returns `b64_json` only, so decode the Base64 before saving. The [GPT Image 2.5 guide](/en/api-capabilities/gpt-image-2-5.md) has a save script.
* `gpt-image-2.5-web` cannot pick Flare or Sunburst.
* Do not reuse every VIP parameter as is; check the web route section of the [GPT Image 2.5 guide](/en/api-capabilities/gpt-image-2-5.md).

```bash theme={null}
export LAOZHANG_API_KEY="your LaoZhang API key"

curl https://api.laozhang.ai/v1/images/generations \
  -H "Authorization: Bearer $LAOZHANG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.5-web",
    "prompt": "An orange cat sitting on a windowsill, afternoon sun, film look",
    "size": "1536x1024"
  }'
```

### Option 2: official forwarding

Official forwarding sends your request on to the official API, so endpoints and parameters match OpenAI's. You can pin Flare or Sunburst, which suits production work that needs consistent output.

1. Create a token under [Token management](https://api.laozhang.ai/token) in the `Sora2Official` or `GPTImage2 Sora2 Enterprise` group, with the billing mode set to Usage first or Usage-based. A `default` token never takes the official-forward route.
2. Set `model` to `gpt-image-2`, `gpt-image-2.5-flare`, or `gpt-image-2.5-sunburst`.
3. Official forwarding is billed per token, so the cost of an image depends on its size and quality. See [GPT Image 2.5 official-forward billing](/en/api-capabilities/gpt-image-2-5.md#official-forward-billing) and the [GPT Image 2 guide](/en/api-capabilities/gpt-image-2.md).

### Option 3: Nano Banana

Nano Banana is billed per call, at the same price from 1K to 4K:

| Model | Model ID | Price |
| - | - | - |
| Nano Banana 2.1 | `gemini-nano-banana-2.1` | \$0.045 per call |
| Nano Banana 2 | `gemini-3.1-flash-image` | \$0.055 per call |
| Nano Banana Pro | `gemini-3-pro-image` | \$0.09 per call |

The request format differs from GPT Image, so changing `model` alone is not enough. 2K and 4K output needs the native Gemini `generateContent` API; the OpenAI-compatible `/v1/chat/completions` returns 1K images. Expect a different visual style, too. See the [Nano Banana 2.1](/en/api-capabilities/nano-banana-2-1-image.md), [Nano Banana 2](/en/api-capabilities/nano-banana2-image.md), and [Nano Banana Pro](/en/api-capabilities/nano-banana-pro-image.md) guides.

## Who is affected

* **Affected**: generation and editing requests to `gpt-image-2-vip`, `gpt-image-2.5-flare-vip`, and `gpt-image-2.5-sunburst-vip`.
* **Not affected**: the web route (`gpt-image-2-web`, `gpt-image-2.5-web`), GPT Image models in the official-forward groups, and the Nano Banana models.

## Check the switch

1. Send a few test requests, open the saved images, and check the size and content.
2. Confirm the model, group, and charges in the [usage log](https://api.laozhang.ai/log).
3. Then move the rest of your traffic. If you stay on `-vip` for now, retry failed requests with backoff and lower your concurrency.

## FAQ

### Will the -vip models be removed?

Not for now; the model IDs still work. We will update this page and post on the [announcements page](/en/changelog.md) once they are stable again.

### Does the web route cost the same as -vip?

Yes. Both are \$0.03 per call, for generation and editing alike. The differences are in parameters and output: the web route returns `b64_json` only, and `gpt-image-2.5-web` cannot pick Flare or Sunburst.

### I need full 2K or 4K detail. Which option?

Use official forwarding (option 2), or Nano Banana through the native Gemini API (option 3). The web route can return large sizes, but official forwarding is the better fit when you need native high-resolution detail.

### I switched to official forwarding but was still billed per call. Why?

The route follows the token's group, not `model`. Check that the key belongs to `Sora2Official` or `GPTImage2 Sora2 Enterprise` and uses a usage-based billing mode.

## Related pages

* [GPT Image 2.5 guide](/en/api-capabilities/gpt-image-2-5.md): parameters, billing, and examples for official forwarding, `-vip`, and the web route
* [GPT Image 2 guide](/en/api-capabilities/gpt-image-2.md): how `gpt-image-2` is billed in each token group
* [Models and pricing](/en/models.md), [usage log](https://api.laozhang.ai/log)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.