Frontier Models
OpenAI Launches GPT Image 2.5 Flare and Sunburst Models
The September 8, 2026 release splits the flagship image system into speed-optimized Flare and precision-optimized Sunburst variants that lead separate leaderboards while sustaining more than 3 billion weekly generations at up to 50 percent lower latency and prior pricing levels.
GPT Image 2.5 is OpenAI's updated image generation and editing model family split into the speed-focused Flare variant and the precision-focused Sunburst variant.
OpenAI released GPT Image 2.5 Flare and GPT Image 2.5 Sunburst on September 8, 2026, as part of the ChatGPT Images 2.5 update. The two models address distinct user needs in text-to-image generation and image editing while rolling out to all ChatGPT, ChatGPT Work, and Codex users on desktop, mobile, and web platforms.
Flare serves as the default model and delivers higher-quality images than the prior GPT Image 2 version at up to 50 percent lower latency. Sunburst targets detailed creative and editing workflows and accepts longer generation times in exchange for added precision in composition and text or symbol edits.
What background context surrounds the GPT Image 2.5 release?
The release continues OpenAI's pattern of iterating on image capabilities through the GPT Image series. Earlier versions established baseline performance in natural language to visual output, yet user feedback highlighted trade-offs between generation speed and edit accuracy that the new split models aim to resolve.
Images 2.5 incorporates gains in natural lighting, richer textures, subject preservation from reference photos, and consistency across multi-turn edits. These enhancements apply to both variants and support professional use cases in film, user-generated content, and advertising.
What details define the GPT Image 2.5 Flare and Sunburst release?
Both models became available through the OpenAI API at the same time as the ChatGPT rollout. Developers access them under the identifiers gpt-image-2.5-flare and gpt-image-2.5-sunburst. The models accept text and image inputs for generation and editing tasks.
Pricing remains identical to the GPT Image 2 tier: five dollars per million text input tokens with one dollar twenty-five cents for cached tokens, eight dollars per million image input tokens with two dollars for cached tokens, and thirty dollars per million image output tokens.
What technical specifics distinguish the two variants?
Flare prioritizes reduced latency while preserving quality improvements over the prior release. Sunburst allocates additional compute to achieve superior results on editing benchmarks, particularly in maintaining character, composition, and visual identity during iterative changes.
On the PiazBench and Artificial Analysis text-to-image leaderboard, Flare holds the first position and Sunburst holds the second. On Arena image editing leaderboards, Sunburst holds the first position and Flare holds the second according to preliminary scores.
| Model | Primary Focus | Latency Profile | Text-to-Image Rank | Image Editing Rank |
|---|---|---|---|---|
| GPT-Image-2.5 Flare | Speed and default use | Up to 50 percent lower than GPT Image 2 | #1 on PiazBench/Artificial Analysis | #2 on Arena |
| GPT-Image-2.5 Sunburst | Precision editing | Longer generation times | #2 on PiazBench/Artificial Analysis | #1 on Arena |
How do the models affect market and stakeholder implications?
The dual-model approach allows different user segments to select the variant that matches their workflow constraints. Creators who value rapid iteration can default to Flare, while teams requiring exacting control over edits can select Sunburst without changing pricing or integration methods.
Availability across ChatGPT platforms and the API broadens access for both consumer and enterprise users. The unchanged pricing structure removes cost barriers that might otherwise slow adoption of the quality and speed upgrades.
What expert reactions have emerged following the announcement?
Industry observers note that the leaderboard splits reflect deliberate specialization rather than a single model attempting to optimize all metrics simultaneously. This design choice aligns with observed demand patterns in creative and production pipelines.
What impressed us most about GPT‑Image‑2.5 Flare is how well it understands what not to change. You can make a meaningful edit without losing the character, composition or visual identity of the original image. That’s incredibly important for the way creators and teams actually work across film, UGC and advertising.Axultan Alimkulov, Head of Product at Higgsfield AI
What developments are expected next for OpenAI image models?
Further iterations may refine the balance between the two variants or introduce additional specialized models. Continued scaling of weekly generation volume beyond the current more than 3 billion images will test infrastructure and optimization techniques.
- Natural lighting improvements
- Richer texture rendering
- Enhanced subject preservation from reference photos
- Improved consistency across multi-turn edits
- Expanded availability to additional ChatGPT tiers
The September 8, 2026 launch positions OpenAI to maintain competitive standing in the frontier image model space while addressing distinct performance dimensions through the Flare and Sunburst variants.
Frequently asked
How do GPT Image 2.5 Flare and Sunburst differ in their intended use cases?
Flare serves as the default choice for faster generation with quality gains, while Sunburst targets workflows that require extra precision in editing and detailed creative tasks.
Sources
- OpenAI — OpenAI released GPT Image 2.5 Flare and Sunburst on September 8, 2026, reporting more than 3 billion weekly images created and up to 50 percent latency reduction.
- OpenAI — GPT Image 2.5 Sunburst generates and edits images from text and image inputs for workflows where editing precision matters most, with token rates matching GPT Image 2.