Tencent Hunyuan Releases Hy4 flash: Per-Task Cost Just One-Twentieth of Hy4

Deep News13:21

On the afternoon of November 16, Tencent Hunyuan unveiled its latest language model, Hy4 flash, which carries a total parameter count of 224B (including 64B Engram parameters) while activating only 12B parameters, and supports multimodality along with a 1M context length.

Hy4 flash stands as the most cost-effective Hy model to date: priced at 0.5 yuan per million input tokens, 2 yuan for output, and just 0.02 yuan when cache is hit, marking a substantial reduction compared with Hy4.

According to the company, to achieve extreme cost-effectiveness while balancing intelligence and practicality, the Hunyuan team carried out a series of efforts. During the pre-training stage, Hy4 flash incorporated architectural features such as hybrid attention, LatentMoE, and Engram, bringing its KV cache size down to just 1/24 of Hy3 and 1/6 of Hy4, while extensive scaling ladder experiments ensured the results met expectations.

In the post-training stage, the team obtained strong overall intelligence through self-distillation from Hy4, and then used reinforcement learning to cut end-to-end task output length by 50%. During the evaluation stage, an internal benchmark set called WorkIndex, built on typical WorkBuddy tasks, was constructed and combined with human evaluation, focusing on the quality of deliverables in daily work, single-turn output length, and end-to-end time, rather than public leaderboard scores far removed from everyday work. In the inference stage, methods including megakernel, 4-bit quantization, and dflash were combined for extreme optimization, pushing Hy4 flash's decoding speed beyond 150 tps and enabling large-scale, low-cost deployment and operation.

Ultimately, Hy4 flash slightly outperforms the larger Hy4 preview (770B total parameters, 49B activated parameters) in overall effectiveness, yet at a significantly lower cost.

Based on comprehensive measurement using the internal WorkIndex benchmark set (covering typical WorkBuddy tasks such as PPT creation and revision, document editing, spreadsheet repair, audio and video processing, and visual understanding), the per-task cost of Hy4 flash is only 1/13 of Hy4 preview, 1/20 of Hy4, and 1/8 of DeepSeek-V4.1-Flash-1027 (Max tier, busy-hour pricing).

Hy4 flash has reportedly launched on products including Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima, and Marvis, and will soon be integrated into OpenRouter. Before November 30, users can try it free for a limited time on WorkBuddy, while the Tencent Cloud TokenHub API is available at a limited-time 50% discount.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Comments

We need your insight to fill this gap
Leave a comment