Artificial Analysis推文
阿里 Qwen-Image-3.0-Pro 登顶图像编辑排行榜第六
阿里巴巴发布第三代图像生成模型 Qwen-Image-3.0 系列，其中 Pro 版在 Artificial Analysis 图像编辑排行榜位列第六，文生图排行榜第九，较上代大幅提升。该系列支持 4.5k token 提示词、精确渲染小至 10 像素的文字及 12 种语言，并通过阿里云 Model Studio 提供 API，定价每张 0.03 至 0.075 美元。
译文
阿里巴巴的Qwen-Image-3.0-Pro在Artificial Analysis图像编辑排行榜上位列第6，在文生图排行榜上位列第9，较上一代Qwen Pro模型Qwen-Image-2.0-Pro分别提升了83和48个Elo分。Qwen-Image-3.0是阿里巴巴Qwen图像家族的第三代产品，包含两个模型：Qwen-Image-3.0-Pro，旗舰版，具备基于智能体的提示词重写功能；以及Qwen-Image-3.0，阿里巴巴将其定位为具备相同能力但生成速度更快、面向高吞吐量场景的版本。Qwen将模型的核心能力聚焦于真实感和信息密度：支持最高4.5k token的提示词，可精确渲染小至10像素的文字，并原生支持12种语言。在Artificial Analysis Image Arena中，Qwen-Image-3.0-Pro在图像编辑排行榜上首次亮相即位列第6，仅次于Microsoft的MAI-Image-2.5-Pro和MAI-Image-2.5、Reve 2.1，以及OpenAI的GPT Image 2和GPT Image 1.5。在文生图排行榜上，它首次亮相即位列第9，紧随Google的Nano Banana 2 Lite之后，领先于ByteDance的Seedream 5.0 Pro。这相较于上一代是一次飞跃：Qwen-Image-3.0-Pro在图像编辑方面比Qwen-Image-2.0-Pro高出83个Elo分（从第34位升至第6位），在文生图方面高出48个Elo分（从第12位升至第9位）。速度更快的Qwen-Image-3.0在文生图排行榜上位列第11，在图像编辑排行榜上位列第15，较Qwen-Image-2.0分别提升了超过135个和近100个Elo分。通过阿里云Model Studio的定价为：Qwen-Image-3.0-Pro在1K分辨率下每张图像0.04美元，2K分辨率下每张0.075美元（每千张图像分别为40美元和75美元）；而Qwen-Image-3.0在两种分辨率下均为每张0.03美元（每千张图像30美元）。两个模型均可通过阿里云Model Studio API使用，并可在Qwen Studio中免费试用。祝贺@Alibaba_Qwen发布新模型！以下为Qwen-Image-3.0-Pro与Artificial Analysis Image Arena中其他领先模型的对比 🧵
Alibaba's Qwen-Image-3.0-Pro lands at #6 on the Artificial Analysis Image Editing Leaderboard and #9 on the Text to Image Leaderboard, up 83 and 48 Elo points on the previous Qwen Pro generation Qwen-Image-3.0 is the third generation of Alibaba's Qwen image family, spanning two models: Qwen-Image-3.0-Pro, the flagship with agent-based prompt rewriting, and Qwen-Image-3.0, which Alibaba positions as the same capabilities with faster generation for high-volume use. Qwen centers the model’s capabilities on realism and information density: prompts of up to 4.5k tokens, text rendered precisely at sizes as small as 10 pixels, and native rendering of 12 languages. In the Artificial Analysis Image Arena, Qwen-Image-3.0-Pro debuts at #6 on the Image Editing Leaderboard, behind only Microsoft's MAI-Image-2.5-Pro and MAI-Image-2.5, Reve 2.1, and OpenAI's GPT Image 2 and GPT Image 1.5. On the Text to Image Leaderboard it debuts at #9, just behind Google's Nano Banana 2 Lite and ahead of ByteDance's Seedream 5.0 Pro. This is a leap over the previous generation: Qwen-Image-3.0-Pro sits 83 Elo points above Qwen-Image-2.0-Pro in Image Editing (#34 to #6) and 48 points above it in Text to Image (#12 to #9). The faster Qwen-Image-3.0 lands at #11 in Text to Image and #15 in Image Editing, up over 135 and nearly 100 Elo points on Qwen-Image-2.0. Pricing via Alibaba Cloud Model Studio is $0.04 per image at 1K resolution for Qwen-Image-3.0-Pro and $0.075 at 2K ($40 and $75 per 1k images), while Qwen-Image-3.0 is $0.03 per image at both resolutions ($30 per 1k images). Both models are available via the Alibaba Cloud Model Studio API and free to try in Qwen Studio. Congratulations to @Alibaba_Qwen on the release! See below for comparisons between Qwen-Image-3.0-Pro and other leading models in the Artificial Analysis Image Arena 🧵