Tencent Hy
Tencent's frontier AI model series
321 followers
Tencent's frontier AI model series
321 followers
Tencent Hy (formerly Hunyuan) is Tencent’s multimodal AI model family, covering text, image, video, 3D generation, and understanding. Built for developers, enterprises, and creators, it powers AI applications, content generation, and next-generation multimodal experiences.
This is the 5th launch from Tencent Hy. View more
Tencent Hy4 flash
Launching today
Hy4 flash is Tencent’s most cost-effective Hy model to date, with 224B total parameters, 12B active, multimodal input and a 1M-token context. Per Tencent, it decodes above 150 tokens per second and costs about $0.07 per million input tokens and $0.28 output. It targets everyday tasks like document work, slides, deep research and computer use, while Hy4 handles harder engineering. Open-sourcing is planned, but the weights are not out yet.






Free Options
Launch tags:Artificial Intelligence
Launch Team

Render Sandboxes: a computer for every agentGive every agent its own machine. Starts in under a second
Promoted

Hy4 flash is the newest model in Tencent’s Hy family, and Tencent calls it the most cost-effective one to date. It’s a smaller, cheaper sibling of Hy4, built for the everyday tasks that make up most AI work.
The specs:
224B total parameters (64B of them “Engram” memory parameters), with 12B active at a time
Multimodal input and a context of up to 1M tokens
Decoding above 150 tokens per second, using 4-bit quantization and other speed-ups
A key-value cache one-sixth the size of Hy4’s, and reinforcement learning that Tencent says cuts average task output tokens by 50%
On Tencent’s own benchmark, called WorkIndex (PowerPoint and document editing, spreadsheet repair, audio and video processing, visual understanding), Hy4 flash slightly outperforms Hy4 Preview, while per-task cost is about 1/13 of Preview’s and 1/20 of Hy4’s. In a blind test with 85 of Tencent’s WorkBuddy users on 259 tasks, satisfaction was close to Claude Haiku 5.5, with the lowest per-task cost and shortest completion time among the models tested.
It isn’t meant to replace Hy4 everywhere. Tencent recommends Hy4 flash for deep research, document processing, slides, simple front-end work and computer use, and points to Hy4 for long-horizon software engineering. Outside write-ups of Tencent’s tests show a wide gap on hard coding benchmarks, for example Terminal-Bench 4.0 at 19.2 for flash against 44.4 for Hy4.
Pricing is about $0.070 per million input tokens, $0.281 output and $0.003 cached input. It’s live on Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis, with OpenRouter coming soon. WorkBuddy use is free through November 30, and TokenHub is at 50% off for a limited time. Tencent plans to open-source Hy4 flash, but the weights and license aren’t out yet.
Good fit if you want a fast, low-cost model for high-volume agent and document workloads, and you’re fine moving to a bigger model for the hardest coding tasks.
P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends