What is GPT-Image-1 Mini?
GPT-Image-1 Mini is OpenAI’s lower-cost image generation and editing model. Its canonical API identifier is gpt-image-1-mini. The model accepts text and image inputs and produces images, so it can support both text-to-image creation and workflows that modify or transform an existing image.
OpenAI released GPT-Image-1 Mini on October 6, 2025, as a more economical alternative to GPT Image 1. Its position in the GPT Image family is practical rather than maximum-quality: it is intended for applications that need many image operations, quick creative exploration, or lower per-image costs.
The model is currently marked deprecated. OpenAI has announced that API access is scheduled to end on December 1, 2026, and lists GPT Image 2 as the recommended replacement. Teams considering GPT-Image-1 Mini for a new production system should therefore treat migration planning as part of the selection decision.
Inputs, outputs, and supported modalities
GPT-Image-1 Mini supports two input types: text and images. Text can describe a new image or specify changes to an existing image. Image inputs can act as source material or references for editing tasks. Its native output is an image.
| Capability | Supported? |
|---|---|
| Text input | Yes |
| Image input | Yes |
| Image output | Yes |
| Text output | No |
| Audio input or output | No |
| Video input or output | No |
This makes the model multimodal in its inputs and outputs, but it is not a general-purpose conversational model. It should be evaluated as an image-generation and image-editing system rather than as a model for producing written answers, audio, video, or structured text responses.
Image quality and size options
Documented generation options include three quality levels: low, medium, and high. The supported image sizes supplied for the model are:
- 1024×1024
- 1024×1536
- 1536×1024
The square format is useful for general-purpose assets, while the other two options provide portrait and landscape orientations. Higher quality and larger output dimensions cost more per generated image. The choice should therefore reflect the stage of the workflow: low quality may be appropriate for rough exploration, while high quality is more suitable when an output is closer to publication or delivery.
The supplied documentation does not specify a general context-window limit or maximum output-token limit. Those values should be treated as unknown rather than assumed from other OpenAI models.
GPT-Image-1 Mini pricing
OpenAI lists both token-based pricing and per-image generation prices. Token pricing separates text content from image content:
| Usage type | Price |
|---|---|
| Text input | $2.00 per 1 million tokens |
| Cached text input | $0.20 per 1 million tokens |
| Image input | $2.50 per 1 million image tokens |
| Cached image input | $0.25 per 1 million image tokens |
| Image output | $8.00 per 1 million image tokens |
OpenAI also publishes per-image generation prices based on quality and dimensions:
| Image size | Low quality | Medium quality | High quality |
|---|---|---|---|
| 1024×1024 | $0.005 | $0.011 | $0.036 |
| 1024×1536 or 1536×1024 | $0.006 | $0.015 | $0.052 |
These prices are especially relevant for high-volume applications. For example, a system generating many low-quality square previews can keep generation costs substantially below a workflow that creates high-quality portrait or landscape images. Actual costs can also reflect text and image input usage, including whether content is eligible for caching.
Strengths and trade-offs
The main strength of GPT-Image-1 Mini is its cost-oriented positioning. It provides image generation and editing capabilities while offering low per-image prices, making it a reasonable fit for workloads where many candidate images are more valuable than the highest possible fidelity on every attempt.
- Lower-cost experimentation: Teams can generate drafts, alternatives, and exploratory concepts without using the highest-priced quality setting for every request.
- Image editing support: Existing images can be used as source material for editing and transformation workflows.
- Flexible output formats: Square, portrait, and landscape sizes cover common creative and product use cases.
- High-volume suitability: The model is designed for repeated generation, previews, and automated variations where per-operation cost matters.
- Input caching: Cached text and image input prices are lower than the corresponding uncached input prices when the workflow qualifies for caching.
The trade-off is that GPT-Image-1 Mini is not positioned as OpenAI’s strongest image model. The supplied research does not provide a benchmark score or quantified quality comparison, so claims about its visual quality should remain qualitative. In practical terms, users who need maximum image quality, complex editing precision, or the most advanced generation performance may be better served by the recommended replacement, GPT Image 2, after validating its price and behavior for their workload.
When to choose this model
GPT-Image-1 Mini is most appropriate when the value of affordable iteration outweighs the need for the highest-fidelity final output. Suitable applications include:
- Generating large batches of early-stage creative concepts
- Producing draft marketing or campaign assets
- Creating preview images before a more expensive final-generation step
- Testing prompts and visual directions rapidly
- Generating automated image variations
- Applying lightweight personalization to images
- Running lower-cost image editing or transformation pipelines
For example, a creative team could use the low or medium quality setting to explore multiple layouts, subjects, or campaign directions, then reserve a higher-quality generation for a selected concept. A software product could also use the model to produce many visual variants for testing or personalization, provided the results meet the product’s quality requirements.
When another option may be more appropriate
GPT-Image-1 Mini may not be the right choice for a workflow that depends on the strongest available image quality, highly precise editing, or a long-lived model with no announced retirement date. It also cannot replace a text, audio, or video model because its native output is limited to images.
The deprecation notice is a particularly important consideration. Although the model remains accessible through the OpenAI API as of September 23, 2026, OpenAI has scheduled API removal for December 1, 2026. New production integrations should assess whether the time available before shutdown justifies the migration work. Existing users should test GPT Image 2 or another approved alternative against representative prompts and source images rather than assuming that outputs, pricing, parameters, or operational limits will be identical.
API status and technical support
GPT-Image-1 Mini is available through OpenAI image-generation and image-editing API surfaces, subject to organization eligibility, rate limits, and any required verification. The canonical model ID is gpt-image-1-mini.
The supplied specifications do not identify tool or function calling, streaming, reasoning, or coding capabilities for this model. Tool use and streaming are recorded as unsupported, while reasoning and coding scores are not provided. Batch API support and caching are recorded in the supplied model data, but these features should still be checked against the current API documentation and the specific endpoint being used.
Because this is an image model, conventional language-model comparisons such as coding benchmarks, text reasoning scores, and text completion limits are not the primary way to evaluate it. More useful tests include image quality at each supported setting, edit consistency, prompt adherence, latency, cost per successful asset, and the amount of human review required.
Is GPT-Image-1 Mini still worth considering?
GPT-Image-1 Mini is a clearly defined cost-and-throughput option for image generation and editing. Its low per-image prices, multiple quality levels, supported portrait and landscape sizes, and image-input capability make it useful for previews, ideation, variations, and lightweight personalization.
Its scheduled shutdown changes the recommendation. The model can make sense for a short-lived experiment or an existing workload that needs economical image operations before migration, but it is a poor choice for a new long-term integration unless the team has a concrete transition plan. For quality-critical work, advanced editing, or a system expected to operate beyond December 1, 2026, evaluate GPT Image 2 or another currently supported image-generation option instead.