Tencent Launches Low-Cost Hunyuan Model Hy4 Flash at One-Twelfth the Input Price of Hy4

Tencent Launches Low-Cost Hunyuan Model Hy4 Flash at One-Twelfth the Input Price of Hy4

Chinese technology company Tencent Holdings has started selling a low-cost version of its Hunyuan model family. Hy4 Flash is now listed on Tencent Cloud's TokenHub API platform at RMB 0.5 per million input tokens and RMB 2 per million output tokens, according to TokenHub's pricing page, last updated on Nov. 16. Cached input is listed at RMB 0.02 per million tokens.

Those list prices are one-twelfth of Hy4's input rate and one-ninth of its output rate on the same official menu, where the flagship model is still priced at RMB 6 and RMB 18 per million tokens. TokenHub also says Hy4 Flash calls will be billed at 50% of the list price through noon Beijing time on Nov. 30. Hy4 itself remains on a separate half-price promotion through Dec. 31.

The cheaper model is already in Tencent's own workplace product. WorkBuddy, Tencent's AI productivity assistant, said on X that Hy4 Flash is live in the model selector, with faster responses, lower cost per task, and built-in vision. It is free to try for a limited time, after which a daily free quota applies and credits are used. ChinaBiz Insider previously described WorkBuddy as Tencent's main commercial vehicle for putting Hunyuan models into office work.

The launch fills out a split that Tencent had already signaled. When it released the 770-billion-parameter Hy4 model in early November, the company said a Flash variant was coming for users who cared more about cost and speed than peak capability. That cheaper model is now a product customers can buy, not a line in a roadmap. Tencent's English Hy4 research page had not yet been updated on the afternoon of Nov. 16 and still described Flash as planned.

What matters for buyers is the shelf, not the parameter count. On TokenHub, Tencent is selling Hy4 Flash next to its own flagship and next to rival models it also hosts. DeepSeek-V4.1-Flash-1017, offered there as an official supply, is listed at RMB 1 per million input tokens and RMB 4 per million output tokens in off-peak hours, and at RMB 2 and RMB 8 during weekday peak hours, according to the same pricing document. Hy4 Flash's undiscounted list price undercuts that off-peak DeepSeek rate. Whether it wins usage will depend on quality in actual work, which Tencent has not documented with an independent evaluation.

TokenHub's model list, also updated Nov. 16, shows Hy4 Flash under the API name hy4-flash, with a 1-million-token context window, image understanding, function calling, and cached context. Hy4 is listed with the same context window. The earlier Hy4 preview is marked for withdrawal on Dec. 7, which would leave paying users with a two-tier Hunyuan lineup: Hy4 for harder jobs, Flash as the cheaper default. TokenHub's changelog records the Flash addition on Nov. 16, after Hy4 itself was added on Nov. 4.

OpenRouter is not in that first wave. Tencent already sells Hy4 through the U.S. model marketplace, and OpenRouter's public catalog on the afternoon of Nov. 16 still listed tencent/hy4 and tencent/hy4-preview, but not a Flash model. Until that listing appears, overseas developers who reach Hunyuan only through OpenRouter do not yet have the low-cost option that TokenHub is selling in China.

The price cut lands on the same day Alibaba is pushing cheaper access to its newest flagship. ChinaBiz Insider reported earlier Monday that Alibaba had launched Qwen4-Max with lower company-reported task costs, a measure distinct from API token prices. Tencent's move is narrower and easier to check: it added a budget model to a public price list and opened it in a product it already operates. It does not, by itself, say how much Hunyuan revenue Tencent expects, or how many TokenHub customers will switch from Hy4 or from the rival models already on the platform.

How the launch benchmarks should be read

In its November 16 launch announcement, Tencent describes an internal blind test involving 85 WorkBuddy users and 259 tasks. That is company-run evaluation evidence, not an independent comparison across all workloads. A lower token tariff also does not guarantee the same saving per completed task when output length, retries and human review differ.

Updated November 17, 2026: Added Tencent’s primary launch announcement and the scope of its internal evaluation. Prices and availability described above are dated observations; check the linked provider terms before purchasing.

Related Coverage:

Tencent Puts Yuanbao Search on the Open Web, After Opening the Stack to Developers

Tencent Hunyuan Hy4 Adds Vision and Computer Use: What the Tests Show

Subscribe to ChinaBiz Insider

Create a free account to join the conversation on ChinaBiz Insider.
[email protected]
Subscribe