On September 22, Tencent officially released Hy Image3.5 preview, its latest professional image generation model. Compared with its predecessor, Hy Image3.0, the new model achieved a 30% higher win rate in human evaluations while delivering improved generation capabilities at a lower cost.
This combination of performance and cost efficiency reflects Tencent Hy’s growing focus on practical utility and real-world business needs.
In an in-house Good/Same/Bad (GSB) blind evaluation involving hundreds of designers, Hy Image3.5 preview performed on par with Seedream 5.0 Pro and slightly better than Nano-Banana Pro and Qwen-Image-3.0 Pro.
Hy Image3.5 preview has been integrated into a range of Tencent’s AI-native products, including the Yuanbao AI assistant, WorkRally for professional film and television production, OnSolo for drama and game content creation, the Miora creative agent, the WorkBuddy office agent workspace, and the ima AI knowledge assistant. API access is also available through Tencent Cloud. International users can experience the model through Miora and OnSolo.
WorkRally is an industrial-grade AI creation platform designed for premium content production. With Hy Image3.5 preview, the platform gains enhanced support for character sheet generation, precise text layouts, and single-element editing.
On July 22, Tencent announced the full rollout of Miora (miora.design), its first in-house creative agent. Built on an agent architecture with the same foundation as WorkBuddy’s, Miora is positioned as a multimodal agent for creative work. Hy Image3.5 preview will be introduced as one of the image generation models available on Miora, accompanied by a two-week free trial.
OnSolo (onsolo.ai) is an AI-native platform for short drama, animation, and interactive film production. It can turn a single idea into scripts, character sheets, storyboards, and finished video. With Hy Image3.5 preview, OnSolo now offers improved character sheet generation, precise text rendering, and single-element editing. Existing members will receive complimentary access for two weeks, while new users will receive 110 free images upon signing up.
Hy Image3.5 preview supports both text-to-image and image-to-image generation, allowing users to upload up to five reference images at a time. Its capability improvements are concentrated in three areas: text rendering and layout, realism and stylistic expressiveness, and consistency in image editing.
These improvements allow the model to support everyday entertainment and learning needs, including photo retouching, portrait enhancement, and educational cards. Professional applications range from commercial posters and television storyboards to film and television promotional materials, UI design, creative illustration, brand visuals, and e-commerce assets.
Hy Image3.5 preview retains the design approach of Hy Image3.0 while further improving language and visual understanding. It delivers stronger image generation results than its predecessor while lowering the cost of use.
Through the Tencent Cloud API, Hy Image3.5 preview is priced at US$0.024 per 2K image. Users are charged only for output images, making professional image generation available at a fraction of what competing models charge.
International image generation model pricing comparison based on the cost of producing one 2K image.
Since beginning an infrastructure overhaul in February, Tencent Hy has released a major model update every two months on average. By aligning development with business use cases and releasing preview versions ahead of full launches, the team can incorporate real-world feedback into its research and development process. Hy Image3.5 preview was developed through a similar approach, bringing the aesthetics of its generated images more closely in line with preferences identified in commercial applications.
In August, Tencent Hy released and open-sourced Hy4 preview, a model designed for real-world productivity tasks spanning coding, office work, and scientific research. In an internal blind evaluation involving 163 in-house experts and 203 engineering tasks, Hy4 preview received an average score of 2.99 out of 4, compared with 2.92 for GLM 5.3 and 2.94 for Kimi K3.
More recently, Tencent Hy combined its large language model and multimodal teams to form a new foundation models department. The release of Hy Image3.5 preview represents another step in Tencent’s continued development of practical, cost-efficient models.
