Technology

Add as a preferred source on Google

Hy4 Flash Runs an Automated Task for 0.038 Yuan, Tencent Says

Tencent says the model has 224 billion total parameters, with 12 billion active per inference, and supports a 1 million-token context.

Updated 1 min read
An engineer watches presentation slides assemble on a laptop during a software test in Shenzhen, China. / TokenPost.ai
An engineer watches presentation slides assemble on a laptop during a software test in Shenzhen, China. / TokenPost.ai

Tencent reported that Hy4 flash's automated-task cost was 0.038 yuan, far below Hy4's task cost, though the smaller model scored lower than Hy4 in the same comparison.

Tencent's comparison gave the flash model 52.2 points and Hy4 56.2. The company put Hy4's cost for that task at about 24 times as much as flash's.

On the PowerPoint collaboration test, Tencent reported scores of 74 for flash and 78.9 for Hy4, with task costs of 0.105 yuan and 1.714 yuan, respectively. On separate benchmarks, Tencent gave Hy4 flash 19.2 points on Terminal-Bench 4.0 and 16 on ProgramBench; Hy4 scored 44.4 and 30, respectively.

Tencent says Hy4 flash has 224 billion parameters, with 12 billion active per inference. Hy4 has 770 billion parameters in total, with 49 billion active per inference. Tencent describes Hy4 flash as multimodal and gives it a 1 million-token context window.

For each million tokens, Tencent lists Hy4 flash input at 0.5 yuan and output at 2 yuan, with cache-hit tokens costing 0.02 yuan. Those rates are about 8.3%, 11.1% and 6.7% of Hy4's listed prices, respectively. Tencent says a temporary 50% discount on Hy4 narrows the difference.

Tencent previously framed Hy4 flash as the cost-conscious model, a direction EconoTimes covered in its report on how the company planned to use the Hy4 lineup.

Related articles

View all