OpenAI's GPT Image 2.6 brings GIF generation, faster speeds, and two new variants, Sunburst and Flare. Here's what changed and what it costs.
What is GPT Image 2.6? #
GPT Image 2.6 is OpenAI’s latest image generation model, and its headline feature is the ability to generate GIFs, not just static images. It also runs faster than its predecessor, roughly 50% quicker according to OpenAI’s own comparison, and ships in two variants: Sunburst, built for higher quality, and Flare, built for speed. The model is available now in ChatGPT, Codex, and through OpenAI’s API.
TL;DR #
- GIF generation is the standout new capability, letting users produce short animated outputs instead of only single frames, something most image models still can’t do natively.
- Two variants ship side by side , Sunburst for quality-focused generation and Flare for faster, lower-latency output, so developers can pick the tradeoff that fits their use case.
- Generation speed improved by about 50% compared to the prior GPT image model, based on OpenAI’s comparison shown alongside the release.
- Pricing is split into input and output tokens , at $8 per million input tokens, an unusual structure for an image model since image generation typically isn’t billed in traditional text-style tokens.
- Availability spans ChatGPT, Codex, and the API , meaning both casual users and developers building on top of OpenAI’s platform get access at launch.
- OpenAI may currently hold the strongest image model among major labs , with competitors like Google focusing more on video generation and ByteDance pushing hyperrealistic stills rather than animated output.
What’s new in GPT Image 2.6? #
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
The two biggest changes are GIF support and speed. Previous GPT image models, like most image generators from any lab, could only output static pictures. GPT Image 2.6 adds the ability to generate short animated GIFs directly, which opens up use cases like simple motion graphics, reaction-style content, or lightweight animated assets without needing a separate video model.
Speed is the second major upgrade. OpenAI’s own comparison shows GPT Image 2.6 generating images about 50% faster than its predecessor. For anyone building a product on top of the API, that kind of latency drop matters as much as raw quality, since image generation speed directly affects how usable a feature feels in a live app.
How do Sunburst and Flare differ? #
GPT Image 2.6 isn’t a single model, it’s released as two variants aimed at different priorities:
- Sunburst targets higher quality output, for cases where fidelity matters more than turnaround time.
- Flare targets speed, trading some quality for faster generation.
This mirrors a pattern increasingly common across AI labs: rather than forcing one model to serve every use case, companies ship a slower “high quality” tier and a faster “lightweight” tier, letting developers choose per request. If you’re generating thumbnails or quick previews, Flare likely makes more sense. If you’re producing final assets for a product or marketing use, Sunburst is the better fit.
What does GPT Image 2.6 cost? #
OpenAI priced GPT Image 2.6 at $8 per million input tokens. The input/output token structure is notable because image models traditionally don’t bill the way text models do, there’s no obvious “token” unit for pixels the way there is for words. That OpenAI applied a token-based pricing scheme to an image model suggests they’re treating image generation more like a unified multimodal pipeline internally, where image data gets tokenized similarly to text. Exact output pricing and per-image cost comparisons between Sunburst and Flare weren’t detailed in the release coverage, so developers evaluating cost at scale should check OpenAI’s own pricing page for the full breakdown before building against it.
Where can you use GPT Image 2.6? #
The model is live in three places:
- ChatGPT , for everyday users generating images (and now GIFs) through the chat interface.
- Codex , OpenAI’s coding-focused product, where image generation can support things like UI mockups or visual assets inside a development workflow.
- The API , for developers building their own applications on top of the model.
Having it available across all three surfaces at launch, rather than staggering availability, suggests OpenAI sees this as a core capability update rather than a niche experiment.
Is GPT Image 2.6 the best image model right now? #
It’s a strong contender, but “best” depends on what you’re optimizing for. GPT Image 2.6’s distinguishing feature is GIF generation, a capability other major labs haven’t matched in their flagship image models. On the other hand, ByteDance’s Seed Dream 5.5 has drawn attention for extremely realistic still images, with fine detail on things like hair texture, skin, and fabric that can look close to photographic. Google, meanwhile, has leaned into video generation with models like Gemini Omni, rather than competing head-on in static image quality, and has reportedly discontinued its Sora-style video effort on the OpenAI side (OpenAI shut down Sora as a standalone product), leaving video generation as more of a Google strength at the moment.
regulatory compliance and security.
So the honest answer is that GPT Image 2.6 is likely the most well-rounded image model among the frontier labs, especially if animated output matters to your use case, but ByteDance’s model may edge it out on raw photorealism for still images. If your priority is video rather than images or GIFs, Google’s offering is currently the more mature option.
Why does GIF support matter for developers? #
For teams building on image APIs, native GIF generation removes a step that previously required stitching together multiple still frames or routing through a separate video pipeline. It’s a smaller lift than full video generation but covers a lot of practical use cases: short looping animations for web content, simple UI feedback, lightweight marketing assets, or social media content that needs motion without the cost and complexity of generating full video clips. Combined with the roughly 50% speed improvement, GPT Image 2.6 positions itself as a faster, more flexible tool for teams that need visual content pipelines but don’t need full video generation.
Frequently Asked Questions #
What is the difference between Sunburst and Flare in GPT Image 2.6?
Sunburst is the higher-quality variant, intended for cases where output fidelity matters more than speed. Flare is optimized for faster generation, trading some quality for quicker turnaround. Developers choose between them based on whether a given task prioritizes speed or image quality.
Can GPT Image 2.6 generate GIFs?
Yes. GIF generation is the headline new feature of GPT Image 2.6, allowing users to produce short animated outputs rather than only static images, a capability not widely available in other frontier image models at the time of release.
How much faster is GPT Image 2.6 than the previous model?
OpenAI’s own comparison shows roughly a 50% speed improvement in image generation compared to its predecessor.
Where can I access GPT Image 2.6?
It’s available in ChatGPT, in Codex, and through OpenAI’s API, covering both consumer and developer use cases at launch.
How is GPT Image 2.6 priced?
OpenAI uses an input/output token pricing model, with input priced at $8 per million tokens. This token-based structure is unusual for an image model, since image generation isn’t typically billed the way text tokens are, but it’s how OpenAI has structured pricing for this release.