The Lab Notes

Tencent

Tencent prices Hy4 flash at $0.070 per 1M input tokens

Tencent released Hy4 flash, its cheapest Hy model, at about $0.070 per 1M input tokens and $0.281 per 1M output, free on WorkBuddy through Nov. 30.

A realistic brass stopwatch with a white face sits at the center of a diagonal split. The left half is dark navy fading to black, crossed by thin white curved lines, and the right half is red-orange, tiled with flat orange stopwatches with lavender crowns linked by blue straps.

Tencent has released Hy4 flash, a model it calls the most cost-effective in its Hy family, at about $0.070 per 1M input tokens. It is live on seven Tencent products, and the weights are still a plan.

Hy4 Flash Is Live on Seven Tencent Products

Tencent released Hy4 flash on Nov. 16, according to its press release. It runs on Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis. Tencent said OpenRouter integration is “coming soon.”

The weights are not out. Tencent said it “plans to open-source Hy4 flash” and gave no date. On Nov. 4, Tencent said it planned to release the weights of full Hy4 in the coming days.

The Price Is $0.070 Input and $0.281 Output

Tencent listed Hy4 flash on Nov. 16 at about $0.070 per 1M input tokens and $0.281 per 1M output tokens, with cached input at $0.003 per 1M. Tencent called that “a substantial reduction from Hy4.”

Through Nov. 30, Hy4 flash is free on WorkBuddy, Tencent’s work assistant app, and 50% off standard pricing on TokenHub.

OpenRouter listed full Hy4 at $0.417 per 1M input tokens and $1.25 per 1M output on Nov. 16. Tencent’s flash price is about 83% lower on input and 78% lower on output, though the two figures come from different sellers.

Hy4 Flash Activates 12 Billion of 224 Billion Parameters

Hy4 flash has 224 billion total parameters, with 12 billion active per token. It takes multimodal input and has a context window of up to 1M tokens.

Tencent said its KV cache, the memory a model holds for the text it has already read, is 1/24 the size of Hy3’s and 1/6 of Hy4’s. The company said decoding runs above 150 tokens per second.

Tencent Says Hy4 Flash Slightly Outperforms a Larger Model on Its Own Benchmark

Tencent said on Nov. 16 that Hy4 flash slightly outperforms the larger Hy4 Preview in overall effectiveness on WorkIndex, an internal benchmark of typical WorkBuddy tasks such as slide editing and spreadsheet repair. Per-task cost on WorkIndex is 1/13 of Hy4 Preview, 1/20 of Hy4 and 1/8 of DeepSeek-V4.1-Flash-1017 at its Max tier with peak-hour pricing, the company said.

In a Nov. 16 blind test with 85 internal WorkBuddy users and 259 tasks in 27 categories, Tencent said Hy4 flash reached user satisfaction close to Claude Haiku 5.5, with the lowest per-task cost and shortest completion time of the models tested. Tencent ran both tests itself.

Tencent Sends Hard Engineering Work to Full Hy4

Tencent recommends Hy4 flash for most everyday tasks, including deep research, document processing, PowerPoint deck generation, simple front-end displays and computer use. For complex work such as long-horizon software engineering, it points users to the full Hy4.

Our View

We think the routing advice matters more than the price cut. Tencent is telling users to run two models, the cheap one for everyday work and the larger one for hard jobs.

For developers, that means splitting traffic between two models. Slide generation, document edits and research can go to Hy4 flash, and long software engineering tasks stay on Hy4. The saving then depends on how much of a team’s volume is everyday work, since only that share moves to the cheaper rate.

WorkBuddy users can try it free through Nov. 30 on their usual decks and spreadsheets.

Every number behind the routing advice comes from Tencent’s own WorkIndex and its own blind test. The split saves money only if flash holds up on a team’s own tasks as well as it did on Tencent’s.