Introducing GPT-Image-2.5 in the API

By OpenAI

Categories: AI, Product

Summary

OpenAI launches two specialized image models: Sunburst for high-fidelity detail work with improved instruction-following, and Flare at 50% faster speed with equal quality. This dual-model strategy lets builders choose between capability and velocity based on workflow needs.

Key Takeaways

  1. Sunburst achieves superior visual fidelity with sharper details, natural lighting, and richer textures while following granular instructions (e.g., which hand someone writes with), enabling multi-iteration refinement for exploratory design workflows.
  2. Flare delivers 50%+ speed improvement over GPT Image 2 without quality degradation, enabling rapid prototyping and exploration of multiple design variations in single sessions.
  3. Both models support transparent backgrounds natively, expanding use cases for asset generation across infographics, game development, and UI/UX design without post-processing.
  4. Dual-model release strategy optimizes for two distinct builder personas: detail-focused creative professionals use Sunburst; speed-focused iterative teams use Flare—available across API, ChatGPT, and Codex simultaneously.
  5. Multi-edit refinement workflow allows preservation of chosen details across iterations, reducing the need for full regeneration and improving design feedback cycles for collaborative projects.

Related topics

Transcript Excerpt

Today, we're launching two new image generation models. GPT Image 2.5 Sunburst and Flare. Sunburst is our most capable model for image generation. It brings significant advances in visual fidelity, including sharper detail rendering, more natural lighting, and richer textures. With Sunburst, you can bring high-quality visual ideas to life with stronger style and detail. You can also refine an idea over multiple edits, like exploring different setups for your dream living room, with better preservation of the details you want to keep. This model follows detailed instructions more closely, down to which hand someone writes with. The second model, Flare, is our fastest image generation model to date. It's over 50% faster than GPT Image 2, while delivering equal quality, so you can explore mor…

More from OpenAI