Kivorent News

Wednesday, November 18, 2026

Tech and AINews

Tencent launches Hy4 Flash at $0.07 per million input tokens

Tencent on November 16 released Hy4 Flash, a smaller, cheaper version of its Hy4 model with 12 billion active parameters and a 1-million-token context window. Its list prices are about 90% below full Hy4, the latest sign of falling prices for everyday AI work.

Cortex, covering AI models, labs and agents
, 3 min read

$0.07Hy4 Flash price per million input tokens

What to know

  • Hy4 Flash has 224 billion total parameters, of which 12 billion are active for each token, and accepts up to 1 million tokens of context.
  • It costs $0.07 per million input tokens and $0.281 per million output tokens on Tencent Cloud, versus $0.834 and $2.501 for full Hy4 in Tencent's Singapore price table.
  • It is free in Tencent's WorkBuddy app and half price on its TokenHub API until November 30; Tencent says it plans to release the weights but gave no date.
List price per million tokens, Tencent Cloud (USD)
Hy4 Flash input$0.07
Hy4 input$0.83
Hy4 Flash output$0.28
Hy4 output$2.5

Source: Tencent Cloud pricing tables as cited by CASRAI, Nov 16, 2026

Tencent released Hy4 Flash on November 16, a smaller and much cheaper version of the Hy4 model it introduced on November 4. The new model is aimed at everyday office and agent tasks, and Tencent priced it at $0.07 per million input tokens and $0.281 per million output tokens, according to Tencent Cloud documentation cited by CASRAI and reports from Tbreak, Superpower Daily, The AI Journal and Inside AI.

What the model is

Hy4 Flash is a mixture-of-experts model with 224 billion total parameters, 64 billion of them in what Tencent calls Engram memory, and 12 billion active for each token, Tbreak and Superpower Daily reported. It reads images as well as text and handles up to 1 million tokens of context, with a maximum of 960,000 tokens in and 64,000 out, per Tencent Cloud's model list as cited by CASRAI.

Tencent says it distilled the model from Hy4 and then used reinforcement learning to cut the length of its answers by half, which lowers what each task costs. It also says the model decodes at more than 150 tokens per second and needs about one-sixth of Hy4's key-value cache memory. These are Tencent's own figures.

Price

On Tencent Cloud, Hy4 Flash costs $0.07 per million input tokens, $0.281 per million output tokens and $0.003 per million cached input tokens. CASRAI cites Tencent's Singapore price table for full Hy4 at $0.834 for input and $2.501 for output, which makes Flash about 92% cheaper on input and 89% cheaper on output. Tbreak listed slightly higher Hy4 prices of $0.90 and $2.70; we used the figures from Tencent's table as CASRAI reported them. For how token prices work, see our explainer on per-million-token pricing.

Tencent separately claims a larger saving per completed task. On its internal WorkIndex test, Hy4 Flash scored slightly above the Hy4 Preview model while costing one-thirteenth as much per task, and one-twentieth as much as full Hy4, the reports said. Inside AI described that as a 95% cut. In a blind test with 85 WorkBuddy users across 259 tasks, Tencent says satisfaction was close to Anthropic's Claude Haiku 5.5. No independent evaluation has been published, and benchmark results run by the vendor deserve caution; see our explainer on what benchmark scores tell you.

Where it runs

The model is available through Tencent Cloud's TokenHub under the name hy4-flash, which accepts OpenAI Chat Completions, OpenAI Responses and Anthropic Messages request formats, according to CASRAI. It also runs inside Tencent's apps, including WorkBuddy, Yuanbao, WeCom and QQ. It is free in WorkBuddy and half price on TokenHub until November 30, Superpower Daily reported. Listing on OpenRouter is planned but not live.

Tencent says it plans to release downloadable weights, but has not given a date, so for now Hy4 Flash is a closed model. Our guide to open-weight and closed models explains the difference.

Why it matters

Hy4 Flash adds Tencent to a price fight among Chinese labs for cheap, long-context models that run agent tasks all day. Its input price sits below the $0.15 that The AI Journal lists for DeepSeek's V4.1-Flash, and it arrives the same day Alibaba made its flagship Qwen4-Max generally available. For developers, that means lower costs for routine work, as long as Tencent's own performance claims hold up in outside testing. DeepSeek recently opened its app tools to API developers in the same market.

Sources

AI modelsAI agents

Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.