The Pulse
Tencent releases Hy4 flash with lower API prices than Hy4
Tencent has released Hy4 flash, a multimodal AI model priced below Hy4 for API use. The company says the model cost less per task than Hy4 and Hy4 Preview in its internal tests.

Tencent has released Hy4 flash, a multimodal AI model with lower API prices than Hy4 for everyday work. The model is priced 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, the company said in its November 16 announcement.
Hy4 flash has 224 billion total parameters, including 64 billion Engram parameters, with 12 billion active parameters. The model supports multimodal inputs and a context length of up to 1 million tokens.
Where it is available
Hy4 flash is available through Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis. Integration with OpenRouter is coming soon.
Through November 30, users can access the model for free on WorkBuddy and receive 50 percent off standard pricing through Tencent Cloud TokenHub. Tencent also plans to open-source Hy4 flash and make frequent iterations to improve it.
Internal tests
Tencent says Hy4 flash slightly outperforms the larger Hy4 Preview in overall effectiveness at a significantly lower cost. Hy4 Preview has 770 billion total parameters and 49 billion active parameters.
On the internal WorkIndex benchmark, the company reports that Hy4 flash cost 1/13 as much per task as Hy4 Preview, 1/20 as much as Hy4 and 1/8 as much as DeepSeek-V4.1-Flash-1017 at its Max tier and peak-hour pricing. WorkIndex consists of eight internal evaluation sets covering tasks such as slide creation and editing, document editing, spreadsheet repair, audio and video processing, and visual understanding.
The benchmark was built around typical WorkBuddy tasks and paired with human evaluations. It focuses on the quality of deliverables, single-turn output length and task turnaround time.
In a separate blind evaluation, 85 internal Tencent WorkBuddy users assessed 259 tasks across 27 categories. Tencent says the model achieved user satisfaction close to Claude Haiku 5.5, with significantly lower per-task cost among the models evaluated and the shortest task completion time.
Model design and intended tasks
Hy4 flash incorporates hybrid attention, LatentMoE and Engram architectural features. Tencent says its KV cache size is 1/24 that of Hy3 and 1/6 that of Hy4.
The company says reinforcement learning reduces average end-to-end task output tokens by 50 percent. It also reports decoding speeds exceeding 150 tokens per second using megakernel optimization, 4-bit quantization and DFly techniques.
The team recommends Hy4 flash for everyday tasks including deep research, document processing, PowerPoint deck generation, simple front-end displays and computer use. For complex tasks such as long-horizon software engineering, it directs users to Hy4.
Source
Companies in this story
Related stories
- Tencent Brings Hunyuan Hy4 to Cloud ADP With Visual Understanding and Computer Operation
- OpenAI Cuts Prices to Win Replit and Rippling Work From Anthropic
- AWS Adds Usage-Based Billing for SageMaker Streaming Models
- Microsoft to bring MAI Cyber 2 Flash to its security products
- Baidu previews ERNIE 5.5 Flash with 1M-token context