GPT-Image-1

On April 23, 2025, OpenAI introduced gpt-image-1 to the API, the natively multimodal model that powered ChatGPT's image generation and had produced over 700 million images for 130 million users in its first week. The model creates images across diverse styles, follows custom guidelines, leverages world knowledge, and accurately renders text for professional-grade output. Leading partners including Adobe (Firefly and Express), Canva (Canva AI and Magic Studio), and GoDaddy (GoDaddy Airo) were already integrating gpt-image-1 for design, logo creation, and editing workflows. The API release gave developers and businesses direct access to the same image generation capabilities used inside ChatGPT, enabling integration into their own tools and platforms across creative, e-commerce, education, enterprise software, and gaming domains. Early use cases highlighted in the announcement include transforming rough sketches into graphic elements in Canva, generating editable logos with background removal in GoDaddy, and offering creators flexible aesthetic style choices through Adobe's ecosystem. The announcement positioned gpt-image-1 as unlocking practical, high-fidelity image generation for production applications.

OpenAI's developer documentation describes GPT-Image-1 as a natively multimodal language model accepting text and image inputs and producing image outputs, with a performance tier labeled "High" at the slowest speed. Pricing is token-based with text input at $5.00 per million tokens ($1.25 cached) and image input at $10.00 per million tokens ($2.50 cached), plus image output at $40.00 per million tokens. Per-image generation costs range from $0.011 (low quality, 1024×1024) up to $0.25 (high quality, 1536×1024 or 1024×1536). The model is exposed across multiple API endpoints including Image generation (v1/images/generations), Responses, Chat Completions, and Batch, with text and image input modalities and image output only; audio and video are not supported. The docs page currently labels GPT-Image-1 as "Our previous image generation model," indicating it has since been superseded in OpenAI's catalog by a newer image model. The page links to a Playground for hands-on testing of the named model.