Advertise
AI

Tencent releases low-cost Hy4 Flash AI model

Hy4 Flash activates 12 billion of its 224 billion parameters per token and is priced at $0.07 per million input tokens, Tencent says.

Tencent released Hy4 Flash, a low-cost AI model aimed at everyday work tasks, on November 16, 2026. According to aitimes.com, the model uses a mixture-of-experts architecture with 224 billion total parameters, of which 12 billion are activated per token; 64 billion of the total are Engram parameters.

Hy4 Flash accepts multimodal inputs including images and offers a context length of up to 1 million tokens. API pricing is set at $0.07 per million input tokens and $0.281 per million output tokens, with cached input at $0.003 per million tokens.

Tencent says KV cache size is reduced to one-sixth that of Hy4 and one twenty-fourth that of Hy3. Reinforcement learning cut the average number of output tokens needed to complete a task by 50%, and the model generates more than 150 tokens per second using 4-bit quantization and inference optimization.

Internal evaluations

In Tencent's WorkIndex evaluation, cost per task was one-thirteenth that of Hy4 Preview, one-twentieth that of Hy4 and one-eighth that of DeepSeek-V4.1-Flash. In a blind test with 85 internal employees across 259 tasks in 27 areas, Hy4 Flash reached user satisfaction similar to Claude Haiku 5.5 and recorded the lowest cost per task and the shortest completion time.

Hy4 Flash is available on Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, Ima and Mavis. Tencent plans to support OpenRouter and to release the model weights in the future.

Tencent unveiled the larger Hy4 model on November 4, 2026, focused on complex reasoning and long software development tasks. Tencent says Hy4 ranked second among Chinese models on Artificial Analysis' AAII intelligence index and first among open-source models on Agent Arena.

Quick answers

How much does Hy4 Flash cost through the API?

Tencent prices it at $0.07 per million input tokens and $0.281 per million output tokens, with cached input at $0.003 per million tokens.

Where can I use Hy4 Flash?

It is available on Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, Ima and Mavis. Tencent also plans to support OpenRouter.

What is Hy4 Flash's context length?

The model supports up to 1 million tokens and accepts multimodal inputs including images.

Source