What is GPT-Image-1.5?
GPT-Image-1.5 is an OpenAI image generation and editing model released on December 16, 2025. It is designed to turn written instructions into images and to modify images supplied by the user. Unlike a text-only language model, its primary output is visual content, although the model documentation also lists text output support.
OpenAI positioned GPT-Image-1.5 for both ChatGPT Images and the OpenAI API. The model is now described as a previous image generation model rather than the provider’s preferred choice for new projects. OpenAI recommends GPT-Image-2 for new development and for migrating existing GPT-Image-1.5 workflows.
What GPT-Image-1.5 can do
The model accepts text prompts and image inputs. A prompt can request a completely new image, describe changes to an existing image, or specify details that should remain unchanged during an edit. This makes it suitable for workflows where preserving composition, lighting, faces, logos, or other important visual elements matters.
- Text-to-image generation: create an image from a written description.
- Image editing: transform or revise an image supplied as input.
- Detail preservation: retain important visual elements during iterative edits.
- Branded content: produce graphics with improved consistency for logos and other brand details.
- Dense text rendering: generate images containing comparatively complex text layouts.
- Quality control: choose low, medium, or high output quality.
- Flexible sizing: use square, portrait, landscape, or automatic sizing options.
These capabilities are particularly relevant when an image must follow several constraints at once. For example, a product team could use the model to create ecommerce imagery while asking it to preserve a product’s shape and branding, or a marketing team could revise a campaign graphic without changing its core composition.
Supported modalities, sizes, and formats
GPT-Image-1.5 supports text and image inputs. It produces image outputs and is also documented as supporting text output. It does not accept audio or video inputs and does not generate audio or video.
| Capability | Support |
|---|---|
| Text input | Supported |
| Image input | Supported |
| Image output | Supported |
| Audio input or output | Not supported |
| Video input or output | Not supported |
| Image quality | Low, medium, and high |
| Documented image sizes | 1024×1024, 1024×1536, 1536×1024, and auto |
| API image formats | PNG, JPEG, and WebP |
The documented size options cover common square, portrait, and landscape use cases. The automatic option can be used when the application does not need to select one of the fixed dimensions explicitly.
GPT-Image-1.5 pricing
OpenAI lists separate token-based prices for text and image processing, along with per-image generation prices. The token prices are relevant when prompts or image data are processed through the API; the per-image prices provide a simpler way to estimate generation costs by quality and resolution.
| Item | Price |
|---|---|
| Text input | $5 per 1 million tokens |
| Cached text input | $1.25 per 1 million tokens |
| Text output | $10 per 1 million tokens |
| Image input | $8 per 1 million tokens |
| Cached image input | $2 per 1 million tokens |
| Image output | $32 per 1 million tokens |
Per-image pricing varies by resolution and quality. For 1024×1024 images, the listed prices are $0.009 for low quality, $0.034 for medium quality, and $0.133 for high quality. For 1024×1536 and 1536×1024 images, the corresponding prices are $0.013, $0.05, and $0.20.
These prices should not be treated as a single flat image-generation rate. A high-quality landscape or portrait image costs more than a low-quality square image, and workflows that include image inputs or text outputs can also incur token-based charges.
Main strengths and trade-offs
GPT-Image-1.5’s most important strength is control over image content rather than general-purpose reasoning. OpenAI’s positioning emphasizes improved instruction following, preservation of important visual details, logo consistency, text rendering, and editing quality compared with GPT-Image-1.
That makes the model a useful fit for production workflows in which an image needs to remain visually coherent across revisions. It is more suitable for a controlled editing pipeline than a workflow that only needs a quick, disposable illustration.
There are also clear trade-offs. High-quality outputs cost substantially more than low-quality outputs, and the model does not provide audio or video generation. It lacks streaming, function calling, structured outputs, and fine-tuning. It also has no documented conventional context-window or maximum-output-token value in the supplied model documentation, because its principal output is image data rather than a conventional text completion.
API support and technical limitations
The canonical API model ID is gpt-image-1.5. The model supports image-generation workflows through OpenAI’s API and is listed as compatible with the Batch API. However, it does not support streaming, function calling, structured outputs, or fine-tuning.
These limitations matter when choosing an implementation pattern. An application can request image generation or editing, but it should not expect the model to behave like an agent that calls external functions, emits structured JSON for downstream systems, or streams a partial image response in the same way a text model might stream tokens.
There is also no published conventional context-length or maximum-output-token specification for this model in the supplied documentation. That absence should be recorded as an undocumented value rather than interpreted as unlimited capacity.
Reasoning, coding, speed, and cost profile
GPT-Image-1.5 is primarily a visual generation model, not a general-purpose reasoning or coding model. The supplied catalog assigns it an editorial reasoning score of 1 and an editorial coding score of 1 on the catalog’s comparative scale. These are editorial evaluations, not OpenAI benchmark results or vendor-published measurements.
The same catalog assigns the model an editorial speed score of 7 and cost score of 6. Those scores are comparative estimates for an image-generation model and should not be read as guaranteed latency or a fixed cost rating. Actual expense depends on input type, image size, quality setting, caching, and the number of images generated.
For applications requiring code generation, complex text reasoning, function calling, or structured data, a dedicated text model is more appropriate. GPT-Image-1.5 should be used when the central deliverable is an image or an image edit.
When to choose GPT-Image-1.5
GPT-Image-1.5 can still be appropriate when an existing application already depends on its behavior and needs a short-term continuation before migration. It is also a reasonable choice for testing image-editing workflows that rely on preservation of visual details, branded graphics, product imagery, or multiple quality levels.
- Choose it for existing production workflows that have not yet migrated.
- Choose it when image editing and preservation of composition or branding are more important than text-model features.
- Choose lower quality for drafts or high-volume exploration where cost matters more than final fidelity.
- Choose medium or high quality when the output is customer-facing or must preserve more detail.
- Choose it for ecommerce, marketing, product catalogs, campaign graphics, and other image-first tasks.
For a new project, GPT-Image-2 is the more appropriate direction because OpenAI explicitly recommends it as the replacement. A text model is preferable for coding, long-form reasoning, structured output, or tool-driven automation. A video or audio model is necessary when the required output is not a still image.
Current status and migration considerations
GPT-Image-1.5 is currently accessible but deprecated. OpenAI has announced that API access will end on December 1, 2026. This makes the model a transitional option rather than a stable long-term foundation for new development.
Teams using the model should identify the exact model ID, image sizes, quality settings, output formats, prompts, and editing assumptions in their current workflow. They should then test those workflows against GPT-Image-2 before the shutdown date. Visual outputs may not be identical after migration, so acceptance tests should compare composition, branding, text rendering, image preservation, and cost at the chosen quality level.
In summary, GPT-Image-1.5 remains useful for image generation and editing, especially where visual preservation and production-oriented controls matter. Its scheduled API removal, lack of agent and structured-output features, and limited modality range mean that new applications should generally use the recommended successor or a different specialized model that better matches their required output.