Tencent Launches Hy4 Flash AI Model at 95% Lower Task Cost

Tencent has launched Hy4 Flash, a lower-cost member of its Hunyuan AI model family, aimed at everyday office work rather than the most demanding AI tasks. The model went live on Monday 16 November 2026, just 12 days after the flagship Hy4 model launched on 4 November.
Hy4 Flash is a 224-billion-parameter model, of which 64 billion parameters are dedicated to what Tencent calls Engram memory. Only 12 billion parameters are activated for any single request, which is how Tencent keeps the cost of running it down. It supports multimodal input, meaning it can work with text and images, and it has a one-million-token context window, allowing it to take in very large documents in a single request.
A smaller, cheaper sibling of Hy4
The flagship Hy4 is considerably larger, with 770 billion total parameters and 49 billion activated per token. Hy4 Flash sits below it as the budget tier of the same family, aimed at routine work such as slide editing, documents, spreadsheets and audio and video processing.
Tencent says a typical task on Hy4 Flash costs about 95 percent less than the same task on the full Hy4 model, based on its own internal benchmark, WorkIndex, which is built from typical tasks in WorkBuddy, its AI work app, combined with human ratings. That benchmark is not public, and the comparison has not been independently verified.
What it costs
API pricing for Hy4 Flash is listed at 0.07 US dollars per million input tokens, 0.281 dollars per million output tokens and 0.003 dollars per million cached input tokens. In Chinese currency, the rates are 0.5 yuan per million input tokens and 2 yuan per million output tokens.
Through 30 November, Hy4 Flash is free inside WorkBuddy and half price on Tencent Cloud's TokenHub API, as a launch promotion. The full Hy4 model is currently on a 50 percent discount, which narrows the real price gap between the two models below the difference in their list prices.
Tencent's performance claims
Tencent says Hy4 Flash performs about as well as Anthropic's Claude Haiku 5.5 on everyday work, citing its own internal tests. In those tests, Hy4 Flash scored 2.79 on user satisfaction, against 2.81 for Claude Haiku 5.5. The company also says the model costs about one-eighteenth as much per task as Claude Haiku 5.5. These are Tencent's own figures, and outside testers have not yet checked them.
Where it is available
The model is already running inside WorkBuddy and Yuanbao, two of Tencent's consumer-facing products, and is available through Tencent Cloud's TokenHub, as well as WeCom, QQ, ima and Marvis. Tencent says OpenRouter is next. The company also says it plans to release Hy4 Flash as an open-weight model, which would let developers download and run it themselves, but it has given no date for that release. The same promise was made for the full Hy4 model, whose weights are still unreleased 12 days after its launch.
Sources: AI Frontier Post and Crypto Briefing.