OpenAI发布GPT-Image-2.6:支持GIF与SVG
OpenAI Launches GPT-Image-2.6 Flare and Sunburst: Adds GIFs, SVG Export and 6-Second API Generation
OpenAI just shipped Images 2.6 across ChatGPT, Codex and the API. For developers, the core release is GPT-Image-2.6, shipped as two API models: Flare and Sunburst. OpenAI team states that Images 2.6 generates images 50% faster than before. The update also adds GIF output, SVG export, markup-guided edits and a new complexity parameter.
OpenAI 刚刚在 ChatGPT、Codex 和 API 中推出了 Images 2.6。对于开发者而言,核心发布版本是 GPT-Image-2.6,以两个 API 模型的形式提供:Flare 和 Sunburst。OpenAI 团队表示,Images 2.6 的图像生成速度比之前快 50%。此次更新还增加了 GIF 输出、SVG 导出、标记引导编辑以及一个新的复杂度参数。
TL;DR
TL;DR(太长不看)
- Size: Closed weights, token-billed image API.
- Runs on: OpenAI-hosted only (ChatGPT, Codex, API).
- Performance: 50% faster image generation than the previous release, per OpenAI.
- Best: Flare can return an image in as little as 6 seconds at low quality.
- Worst: No benchmark scores and no per-image cost table published for 2.6.
- 规模:闭源权重,按 token 计费的图像 API。
- 运行环境:仅限 OpenAI 托管(ChatGPT、Codex、API)。
- 性能:根据 OpenAI 的说法,图像生成速度比上一版本快 50%。
- 最佳表现:Flare 在低质量设置下可在短短 6 秒内返回一张图像。
- 最差表现:未公布 2.6 版本的基准测试分数,也未发布单张图像的成本表。
image source: openai
图片来源:openai
What is GPT-Image-2.6?
什么是 GPT-Image-2.6?
GPT-Image-2.6 is OpenAI’s image generation and editing model family for the API. It accepts text and image inputs and returns images. GPT-Image-2.6 Flare is built for fast iteration and interactive workflows. GPT-Image-2.6 Sunburst is built for detailed compositions and editing.
GPT-Image-2.6 是 OpenAI 面向 API 的图像生成与编辑模型系列。它接受文本和图像输入,并返回图像。GPT-Image-2.6 Flare 专为快速迭代和交互式工作流而构建。GPT-Image-2.6 Sunburst 则专注于精细构图和编辑。
OpenAI states that Flare can generate images in as little as 6 seconds when quality is set to low. Sunburst targets higher quality and stronger consistency across edits at medium or high quality.
OpenAI 表示,当质量设置为低时,Flare 可在短短 6 秒内生成图像。Sunburst 针对中等或高质量设置下的更高画质以及在编辑过程中保持更强的一致性。
What changed in ChatGPT Images 2.6?
ChatGPT Images 2.6 有哪些变化?
ChatGPT users can now turn a prompt or a photo into an animated GIF. They can adjust animation speed, edit individual frames and share the result. Users can also save and tag reference images of people, pets and characters. Those references can be reused later without uploading them again.
ChatGPT 用户现在可以将提示词或照片转换为动画 GIF。他们可以调整动画速度、编辑单个帧并分享结果。用户还可以保存并对人物、宠物和角色的参考图像进行标记。这些参考图像可以在后续使用中重复调用,无需再次上传。
Usage keeps climbing. OpenAI says people now generate more than 4B images per week across ChatGPT and the API. That is up from more than 3B per week 8 weeks earlier, at the Images 2.5 launch. On average, that is over 500M images per day.
使用量持续攀升。OpenAI 表示,目前用户在 ChatGPT 和 API 中每周生成的图像已超过 40 亿张。而在 8 周前 Images 2.5 发布时,这一数字为每周超过 30 亿张。平均每天生成超过 5 亿张图像。
How does the new complexity parameter work?
新的复杂度参数如何工作?
The complexity field sets how much the model plans before it renders. The image generation guide lists 3 levels: low, medium and high. If you omit it, both models use high.
complexity 字段用于设定模型在渲染前的规划程度。图像生成指南列出了三个级别:low(低)、medium(中)和 high(高)。如果省略该参数,两个模型均默认使用 high。
Low suits focused edits, such as changing a background or a sign. High helps with dense text, complex layouts and relationships between objects. OpenAI warns that higher complexity can add latency and introduce unwanted changes. Adjusting complexity has no separate fee. It is separate from quality, which controls rendering detail and fidelity.
Low 适合聚焦式编辑,例如更改背景或标志牌。High 有助于处理密集文本、复杂布局以及对象之间的关系。OpenAI 警告称,更高的复杂度可能会增加延迟并引入不必要的更改。调整复杂度不收取额外费用。它与 quality 是分开的,后者控制渲染的细节和保真度。
Markup edits, SVG and GIF output in the API
API 中的标记编辑、SVG 和 GIF 输出
Image markup lets you mark up the exact region you want changed. The launch post says markup is coming to the API. The guide already lists markup-guided edits as supported for both models. It also lists SVG output and animated GIF output through the Image API.
图像标记功能允许您对需要更改的确切区域进行标记。发布帖子表示标记功能即将推出至 API。指南已将标记引导的编辑列为两种模型均支持的功能。它还列出了通过 Image API 输出 SVG 和动画 GIF 的功能。
SVG matters for icons, diagrams and UI assets that must scale without blur. GIF output removes a separate animation step for social and product teams.
SVG 对于必须无损缩放的图标、图表和 UI 资产至关重要。GIF 输出为社交和产品团队省去了单独的动画制作步骤。
How much does GPT-Image-2.6 cost?
GPT-Image-2.6 的价格是多少?
Pricing is token-based and identical for Flare and Sunburst, per the model pages:
根据模型页面显示,Flare 和 Sunburst 均采用基于令牌的定价,且价格相同:
- Text input: $5.00 per 1M tokens ($1.25 cached)
- Image input: $8.00 per 1M tokens ($2.00 cached)
- Image output: $30.00 per 1M tokens
- 文本输入:每 100 万令牌 5.00 美元(缓存后 1.25 美元)
- 图像输入:每 100 万令牌 8.00 美元(缓存后 2.00 美元)
- 图像输出:每 100 万令牌 30.00 美元
How does GPT-Image-2.6 compare with Nano Banana 2.1 and FLUX 3 Image?
GPT-Image-2.6 与 Nano Banana 2.1 和 FLUX 3 Image 相比如何?
| Feature | GPT-Image-2.6 Flare / Sunburst | GPT-Image-2.5 Flare / Sunburst | Google Nano Banana 2.1 | BFL FLUX 3 Image |
|---|---|---|---|---|
| Released | Nov 4, 2026 | Sep 8, 2026 | Oct 2026 | Oct 1, 2026 |
| Parameters | Not disclosed | Not disclosed | Not disclosed | Not disclosed (Klein: 5.6B) |
| Weights | Closed | Closed | Closed | Closed (Klein open, FLUX Kommunity License) |
| Planning control | complexity: low / medium / high | Not disclosed | Thinking: minimal / medium / high | Bounding boxes in prompt |
| Max reference images | Not disclosed | Not disclosed | 14 | 10 |
| Max output | Not disclosed | Not disclosed | 4K | 4K (about 16 MP) |
| GIF output | Yes | Not disclosed | Not disclosed | Not disclosed |
| SVG output | Yes | Not disclosed | Not disclosed | Not disclosed |
| Pricing | $8 / $30 per 1M image in / out | $8 / $30 per 1M image in / out | $0.0336 per 1K image | $0.048 per 1K image |
| Speed claim | Flare: 6 s minimum at low quality | Flare: up to 50% lower latency vs GPT-Image-2 | Not disclosed | Not disclosed |
| 功能 | GPT-Image-2.6 Flare / Sunburst | GPT-Image-2.5 Flare / Sunburst | Google Nano Banana 2.1 | BFL FLUX 3 Image |
|---|---|---|---|---|
| 发布日期 | 2026 年 11 月 4 日 | 2026 年 9 月 8 日 | 2026 年 10 月 | 2026 年 10 月 1 日 |
| 参数量 | 未披露 | 未披露 | 未披露 | 未披露(Klein: 56 亿) |
| 权重 | 闭源 | 闭源 | 闭源 | 闭源(Klein 开源,采用 FLUX Kommunity 许可证) |
| 规划控制 | 复杂度:低 / 中 / 高 | 未披露 | 思考模式:最小 / 中 / 高 | 提示词中的边界框 |
| 最大参考图像数 | 未披露 | 未披露 | 14 | 10 |
| 最大输出分辨率 | 未披露 | 未披露 | 4K | 4K(约 1600 万像素) |
| GIF 输出 | 支持 | 未披露 | 未披露 | 未披露 |
| SVG 输出 | 支持 | 未披露 | 未披露 | 未披露 |
| 定价 | 每张输入/输出图像每百万张 8 / 30 美元 | 每张输入/输出图像每百万张 8 / 30 美元 | 每张图像每千张 0.0336 美元 | 每张图像每千张 0.048 美元 |
| 速度声明 | Flare:低质量下最低 6 秒 | Flare:延迟较 GPT-Image-2 降低高达 50% | 未披露 | 未披露 |
Google and BFL publish per-image prices. Nano Banana 2.1 starts at $0.0336 per 1K image and reaches $0.113 at 4K. FLUX 3 Image starts at $0.048 per 1K image and reaches $0.607 at 4K. Its open-weight sibling, FLUX 3 Image Klein, runs on consumer GPUs down to an RTX 3090. OpenAI’s edge is native GIF and SVG output. Its gap is transparency on per-image cost and quality.
Google 和 BFL 公布的是单张图片价格。Nano Banana 2.1 起步价为每千张图片 0.0336 美元,在 4K 分辨率下达到 0.113 美元。FLUX 3 Image 起步价为每千张图片 0.048 美元,在 4K 分辨率下达到 0.607 美元。其开源权重的兄弟产品 FLUX 3 Image Klein 可在消费级 GPU 上运行,最低支持 RTX 3090。OpenAI 的优势在于原生支持 GIF 和 SVG 输出。其劣势在于单张图片成本和质量的透明度不足。
Key Takeaways
关键要点
- Flare returns images in as little as 6 seconds at low quality.
- Sunburst targets detail and edit consistency at medium or high quality.
- New complexity parameter (low, medium, high; default high) costs nothing extra.
- Token prices match GPT-Image-2.5: $30 per 1M image output tokens.
- 在低质量设置下,Flare 最快可在 6 秒内返回图像。
- Sunburst 针对中等或高质量下的细节和编辑一致性进行了优化。
- 新增复杂度参数(低、中、高;默认高),不收取额外费用。
- 令牌价格与 GPT-Image-2.5 保持一致:图像输出令牌每百万个 30 美元。
Check out the launch post, the Flare model card, the Sunburst model card and the image generation guide. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
查看发布帖子、Flare 模型卡片、Sunburst 模型卡片以及图像生成指南。此外,欢迎在 Twitter 上关注我们,并别忘了加入拥有超过 15 万成员的 ML SubReddit,同时订阅我们的新闻通讯。等等!你在使用 Telegram 吗?现在你也可以加入我们的 Telegram 群组。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力