Independent evaluator Artificial Analysis benchmarked Apple's ADM 3 Cloud, the model behind iPhone photo editing, in its image arena. It placed 42nd on AA-Image-Editing v2.1 and 85th on AA-Image-T2I v2.1, giving a third-party quantitative read on Apple's image AI: middling for editing and clearly behind leaders on pure text-to-image generation.
Text-to-Image
文生图
Task paradigm of generating images from text prompts, commonly benchmarked for quality and speed.
Curated Intelligence
Artificial Analysis' independent benchmark shows GPT Image 2.6 Flare generating images at a median of about 30 seconds at max quality, nearly twice as fast as GPT Image 2.5 Flare's roughly 55 seconds, while also scoring higher on both AA-Image-T2I v2.1 and AA-Image-Editing v2.1. The update delivers simultaneous gains in speed and quality for image generation.
OpenAI released GPT Image 2.6 on November 4 across ChatGPT, Codex and the API, shipping two variants, Flare and Sunburst. Artificial Analysis benchmarks show both taking the top two spots on AA-Image-T2I v2.1 and AA-Image-Editing v2.1: Flare nearly doubles generation speed with slightly better quality, while Sunburst trades longer generation time for the #1 quality slot.
Black Forest Labs released FLUX 3 Image Klein, a 5.6B-parameter open-weights image model, on October 15. It ranks #2 among open-weights models on Artificial Analysis' AA-Image-T2I v2.1 leaderboard, essentially tied with #1 Qwen-Image-2.1, and #3 on AA-Image-Editing v2.1. Under the FLUX Kommunity License it is free for non-commercial use, with commercial output rights for individuals and companies under $5M annual revenue; API pricing is $0.009 per 1K image, 40% below the 9B FLUX.2 [klein].
Chinese GPU maker Iluvatar CoreX launched the TianGai 300 chip based on its 'Tianshu' architecture, which outperformed Nvidia's H200 in a live text-to-image benchmark (95.6s vs 112s). The company also introduced the edge-side TongYang TY1200, demonstrating superior speed over Jetson Thor T5000 in offline scenarios. This release fulfills previous roadmap promises to surpass the Hopper architecture and includes a new AI development software stack to build a cloud-edge computing ecosystem.
BaoCut 3.5 integrates Qwen Image 2.1 Turbo to enable fully local image generation without cloud dependencies. On a Mac Mini M2, it generates 1080P images in 3-5 minutes and 2K images in nearly 20 minutes. The open-source tool supports direct invocation by AI Agents for automated drawing and has been observed generating NSFW content.
Alibaba's Tongyi Qianwen has officially launched the Qwen-Image-2.1 Pro and Turbo APIs for image generation and editing, supporting natural language refinement and multi-reference control. The Turbo model weights are also open-sourced, allowing developers to integrate via API or deploy locally to streamline creative workflows.
ByteDance's Doubao Work platform introduced an infinite canvas feature for organizing assets and iterative design editing. It also integrated the new Seedream 5.0 Flash image model and launched the Doubao 2.1 Lite model to enhance multimodal creation efficiency in high-frequency scenarios like document and PPT generation.
Google Research introduces the 'Diffusion Controller' framework, aiming to simplify complex control flows in text-to-image models via a unified controller mechanism. By leveraging a steering damper network for precise intervention in the generation process, the approach enhances controllability, mitigates harmful content generation, and shows potential for adaptation to next-generation video models.
Google launches Google Pics, an AI image generation and editing tool integrated directly into Workspace. Users can create, refine, and translate images via text prompts within Docs and Slides without switching apps. The feature is available now, with Drive integration rolling out in coming weeks.