Alibaba Tongyi released the high-speed version of its image model, Qwen-Image-2.1-Turbo, on October 9, supporting image generation and editing in eight denoising steps, with Pro and Turbo API interfaces launched simultaneously. On October 11, the Qwen-Image-2.1 Pro and Turbo APIs officially went live on the Qwen AI platform, allowing developers to directly call APIs for image generation and editing capabilities.
By using low-step denoising to produce images, the model substantially compresses generation time while balancing generation speed with editing capability, providing a lightweight API option for graphic creation, design and content production scenarios.
Zhipu AI, meanwhile, brought out the GLM-5.3-Highspeed model on its Zcode platform, currently in grey release starting with version 3.15.1 and randomly opened to subscribers, with measured speeds reaching up to 490 tokens per second. The model offers a one-million-token context window and a maximum output of about 128,000 tokens, and supports low, high and max reasoning effort configurations. Compared with the previous GLM-5.3-FlashX, which peaked at 200 tokens per second, the gain in generation performance is substantial and long-text output efficiency is markedly improved.
Commentators note that the two-version API for Tongyi's image model enriches the multimodal capability matrix on the Qwen platform, further lowering the cost for enterprises and developers to access high-quality image generation and accelerating multimodal AI adoption across business scenarios.
Source: iiMedia / AITOP100
