OpenAI rolled out two new image generation models under the ChatGPT Images 2.5 banner, but the company remains vague about rollout timing and which users get access to which model. The move splits the image generation stack into two specialized tools: Flare, built for speed, and Sunburst, designed for precision in editing tasks.
Flare prioritizes generation velocity. Tests show the model produces images faster than previous versions, a meaningful improvement for users who generate dozens of variations or iterate rapidly through concepts. The speed gains matter most for creative workflows where latency compounds. A designer mocking up five poster variations back-to-back saves minutes with faster generation times. OpenAI has not published exact benchmarks, but internal testing reportedly shows substantial speed gains across standard prompts.
Sunburst takes the opposite approach. It refines OpenAI's editing capabilities, focusing on inpainting, outpainting, and targeted modifications to existing images. The model produces more precise edits with fewer artifacts around modified regions. Users editing specific areas of an image without regenerating the entire composition report cleaner results and fewer hallucinations at edit boundaries. This addresses a persistent problem in generative image tools: editing often creates visible seams or quality drops in affected regions.
The deployment strategy introduces friction. OpenAI has not clearly communicated which subscription tier receives which model, whether both roll out simultaneously, or if selection happens automatically based on task type. Users report accessing different models unpredictably. Some ChatGPT Plus subscribers report consistent access to Flare for general generation, while access to Sunburst remains sporadic. Free-tier users see even less clarity. This mirrors OpenAI's rollout pattern for previous models, but the dual-model approach adds complexity.
The distinction matters because the models solve different problems. Flare benefits high-volume users who generate many images daily. Content creators, marketers, and designers who iterate rapidly see tangible workflow improvements. Sunburst serves detail-focused work: product mockups, architectural visualizations, or any task requiring surgical edits. A user might need both models for different jobs, but OpenAI has not confirmed whether users can request specific models or if selection occurs automatically.
Performance gaps emerge based on use case. Images 2.5 shows meaningful speed improvements on photorealistic prompts and complex scenes. Abstract or simple prompts see less dramatic gains, suggesting optimization targeted specific image categories. Sunburst's precision gains concentrate on architectural details, fabric textures, and facial features. Blurry backgrounds and simple backgrounds reportedly show minimal improvement, indicating the model trains more heavily on complex, intricate imagery.
The pricing picture remains unresolved. OpenAI charges per image generated or edited through ChatGPT Plus subscriptions. Whether Images 2.5 changes per-generation costs or locks faster generation behind a higher tier remains unknown. This uncertainty affects adoption calculations for creative professionals evaluating ChatGPT as a production tool.
OpenAI's approach reflects a market shift toward specialized models over monolithic systems. Rather than releasing one improved model for all tasks, the company bet on purpose-built alternatives. This strategy succeeds only if users understand the tradeoff and can access both models when needed. Current rollout messaging fails that test.