Skip to content

Tencent ships Hy4 flash, a model 92% cheaper than its own flagship

Hy4 flash cuts input and output pricing by around 90% against full Hy4, while user satisfaction in internal testing came close to Claude Haiku 5.5.

Source: Tencent

Tencent released Hy4 flash on 16 November, a smaller, cheaper sibling to the Hy4 model it shipped earlier this month. It’s a mixture-of-experts model with 224 billion total parameters, 12 billion active per token and 64 billion Engram parameters, with a 1 million-token context window and native multimodal input.

Pricing is $0.070 per million input tokens and $0.281 per million output, with cached input at $0.003. Tencent says that’s roughly 92% and 89% lower than full Hy4’s input and output prices. Decoding speed is over 150 tokens a second. In a blind internal evaluation, 85 WorkBuddy users rated Hy4 flash across 259 tasks in 27 categories, and Tencent says satisfaction came close to Claude Haiku 5.5, with the lowest per-task cost and fastest completion time of the models tested. It’s live now on Tencent Cloud TokenHub and several Tencent apps, with OpenRouter access coming soon, and Tencent plans to open-source it with frequent updates.

Why it matters: a model vendor undercutting its own flagship by roughly 90% on price, while claiming satisfaction close to a known model like Claude Haiku 5.5, is a real signal for anyone routing requests by cost. It’s also a reminder that the fast-tier pattern, where a provider ships a near-free sibling to its frontier model, is now standard across Chinese and Western labs alike.

The caveat: the benchmark is Tencent’s own internal user panel, not a public, reproducible eval, and open weights aren’t out yet.

Share

Related reading

More from The Wire

All briefs