Image: Tencent
KUALA LUMPUR, November 17, 2026 — Tencent has released Hy4 flash, a cheaper and faster version of its flagship Hy4 artificial intelligence model, aimed squarely at the everyday office work — research, documents, slides and spreadsheets — that makes up much of what businesses actually ask AI to do.
The Chinese technology group announced the model on Monday (November 16). According to Tencent, Hy4 flash has 224 billion parameters in total, including 64 billion so-called Engram parameters, but activates only 12 billion of them at a time. That approach lets a large model run with the computing bill of a much smaller one. Like Hy4, it accepts multimodal inputs and supports a context window of up to one million tokens.
Price is the headline. Tencent lists API rates of about US$0.070 per million input tokens, US$0.281 per million output tokens and US$0.003 per million cached input tokens. Its published rates for the full Hy4 are US$0.834 for input and US$2.501 for output per million tokens.
On WorkIndex, an internal benchmark built around typical tasks in its WorkBuddy app, such as creating and editing slide decks, editing documents, repairing spreadsheets, processing audio and video and visual understanding, Tencent said Hy4 flash cost one-twentieth as much per task as Hy4, one-thirteenth as much as the larger Hy4 Preview, and one-eighth as much as DeepSeek-V4.1-Flash-1017 at peak-hour pricing. The company said it also slightly outperforms Hy4 Preview, a model with 770 billion total and 49 billion active parameters, in overall effectiveness.
In a blind evaluation with 85 internal WorkBuddy users across 259 tasks in 27 categories, Tencent said Hy4 flash achieved user satisfaction close to Anthropic’s Claude Haiku 5.5, with the shortest completion time and the lowest per-task cost of the models tested. These are Tencent’s own tests rather than independent benchmarks; TechNode, which also reported the launch, attributed the figures to the company.
How did Tencent get the cost down? It said the model’s memory cache is one-sixth the size of Hy4’s, reinforcement learning cut average output length per task by 50 per cent, and a combination of megakernels, 4-bit quantisation and a technique it calls DFly pushes decoding speeds beyond 150 tokens per second.
Hy4 flash is available through Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis, Tencent said, with OpenRouter integration to follow. Until November 30, WorkBuddy users can use it free, and TokenHub is offering API access at half the standard price. Tencent also plans to release the model as open weights, though it has not given a date.
The company recommends Hy4 flash for most everyday tasks, including deep research, document processing, slide generation, simple front-end demos and computer use, and points users to the full Hy4, released on November 4, for complex jobs such as long-horizon software engineering.
Why it matters: Chinese AI developers are competing hard on price, and Tencent’s launch comes a day after Alibaba unveiled its Qwen4-Max flagship. For businesses in Malaysia and across South-East Asia that build chatbots and automation on pay-per-use AI services, falling token prices change the economics of what is worth automating. An open-weight release would also let companies run the model on their own hardware rather than through Tencent’s cloud.
What to watch next is whether independent testers confirm Tencent’s cost and quality claims, and when the open weights arrive.
Sources: Tencent, November 16, 2026; Tencent Hy; TechNode.

