Tencent Launches Hunyuan Hy4 Flash for High-Speed and Cost-Effective AI Workflows
According to Tencent Newsroom: Tencent reports the release of its Hunyuan Hy4 flash model, featuring a 224B total parameter count with 12B active parameters. Designed for everyday workflows, it offers significant cost reduction and high-speed decoding exceeding 150 tps.

Original image temporarily unavailable
According to Tencent Newsroom:
According to Tencent, the newly released Hy4 flash model features 224B total parameters, including 64B Engram parameters, while keeping active parameters at 12B. It supports multimodal tasks and a 1M context length.
To achieve high efficiency, Tencent states that the model incorporates architectural features like mixed attention, LatentMoE, and Engram during pre-training, which reduces the KV cache size significantly. Through self-distillation and reinforcement learning, the model's end-to-end task output length is reduced by 50%.
In terms of pricing and deployment, Tencent reports that the input price is 0.5 RMB per million tokens and output is 2 RMB, dropping to 0.02 RMB with cache hits. Internal blind tests show that Hy4 flash achieves close user satisfaction to competitive models while maintaining lower task costs and shorter processing times. It is currently rolling out across several Tencent products and services.
View selection rationale
Importance 75/100
About this brief
- Content type
- News summary
- Source type
- First-party / official source
- Original source
- Tencent Newsroom ↗
- Source published
- Process
- AI-assisted discovery, translation, summarisation, classification and scoring; standards and corrections are maintained by the editorial desk.
- Verification
- The original URL and publication metadata are retained. This brief is not independently fact-checked unless explicitly stated.
- Published by
- Polyemergence Editorial Desk · PolyEmergence Pte Ltd.
- Conflict disclosure
- No financial or commercial conflict declared.