Monday, 16 November 2026

Tencent’s Hy4 flash model is priced low and free in WorkBuddy to 30 November

Tencent has priced its new Hy4 flash AI model at $0.07 per million input tokens, with free use in WorkBuddy until 30 November and weights promised.

Key points

  • Hy4 flash costs $0.07 per million input tokens and $0.281 per million output tokens.
  • WorkBuddy users get it free through 30 November.
  • Tencent plans to release weights and add it to OpenRouter.

Tencent wants office users to try a cheaper AI model, and until 30 November it is free in the WorkBuddy app. The model, Hy4 flash, is also inside the Yuanbao chatbot and on Tencent Cloud’s TokenHub, where developers can buy access.

Developers pay $0.07 to send in a million tokens, $0.281 to get a million back, and just $0.003 when the input is already cached. TokenHub is halving those figures for now, giving $0.035 and about $0.14 for the first two. Tencent has not said when that offer ends.

Tencent’s claim is about cost against its bigger sibling. Running the same job on full Hy4 costs about 20 times as much, by the numbers from Tencent’s in-house WorkIndex. Separately, 259 tasks were scored blind by 85 WorkBuddy users; Hy4 flash came out near Claude Haiku 5.5 for satisfaction, and Tencent says it was also the cheapest and fastest of those compared. Both tests are Tencent’s own.

Technically it is a mixture-of-experts network. Of its 224 billion parameters, 12 billion work on any single pass, and it reads images alongside text, with room for a million tokens of context. The jobs Tencent pitches it for are research, slide decks and spreadsheet edits.

For developers outside China, the likely routes are the Tencent Cloud API now and OpenRouter later, since Tencent plans to list it there and to release the weights. No date was given for either, and no international WorkBuddy rollout was mentioned.

Claude Haiku 5.5 is the one rival Tencent names, so the comparison is with a small, fast model rather than a flagship. That is a fair yardstick for a model sold on price, though it leaves open how Hy4 flash handles harder work.

TPRO Team

The Tech Prolonged newsroom desk. Stories under this byline cover AI, the web and general tech, and are reviewed by our editors before publishing.