Tencent Launches Hy4 flash: 224B Parameter Model with 12B Active Parameters at 1/13 Cost of Hy4 Preview
Tencent's new Hy4 flash model uses hybrid attention, LatentMoE and Engram architecture to deliver 150+ tokens per second at a fraction of predecessor costs, now free on WorkBuddy through November 30.

By TechWhoKnows
3 min read
Tencent has launched Hy4 flash, the latest addition to its Hunyuan (Hy) model family, positioning it as the most cost-effective model in the lineup with a 224 billion parameter architecture that activates only 12 billion parameters per forward pass.
Architecture and efficiency
Hy4 flash incorporates hybrid attention, LatentMoE and Engram architectural features. The model includes 64 billion Engram parameters within its 224 billion total. Its KV cache size is just 1/24 that of Hy3 and 1/6 that of Hy4. Extensive scaling ladder experiments were conducted during pre-training to validate performance.
Post-training reinforcement learning cuts average end-to-end task output tokens by 50 percent. Inference optimisations combining megakernel, 4-bit quantisation and DFly techniques achieve decoding speeds exceeding 150 tokens per second.
Pricing and availability
API pricing is set at approximately US$0.070 per million input tokens, US$0.281 per million output tokens, and US$0.003 per million cached input tokens, a substantial reduction from Hy4. Hy4 flash is available across Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis, with OpenRouter integration coming soon. From launch through 30 November, users can access Hy4 flash for free on WorkBuddy.
WorkIndex benchmark and human evaluation
Tencent built WorkIndex, an internal benchmark based on typical WorkBuddy tasks, paired with human evaluations. The benchmark covers PowerPoint deck creation and editing, document editing, spreadsheet repair, audio/video processing and visual understanding tasks. The team focused on deliverable quality, single-turn output length and task turnaround time rather than public benchmark scores.
On WorkIndex, Hy4 flash slightly outperforms the larger Hy4 Preview (770B total parameters, 49B active parameters) in overall effectiveness at significantly lower cost. Per-task cost is reported as 1/13 that of Hy4 Preview, 1/20 that of Hy4, and 1/8 that of DeepSeek-V4.1-Flash-1017 (Max tier, peak-hour pricing).
In a blind evaluation with 85 internal Tencent WorkBuddy users across 259 tasks in 27 categories, Hy4 flash achieved user satisfaction close to Claude Haiku 5.5, with the significantly lower per-task cost among all models evaluated and the shortest task completion time.
Sources
- 1PRNewswire - Tencent announcementprnewswire.com