Flash

Flash·Models & Products·2026-11-16 20:07

Tencent Launches Hy4 Flash: 224B Parameters at 1/20th the Per-Task Cost

Tencent launched Hunyuan Hy4 flash on November 16, a low-cost model for everyday office work: 224 billion total parameters with 12 billion activated per token, multimodal input and a one-million-token context, priced well below the flagship Hy4.

A laptop showing charts and code in a bright office
AI-generated illustration

What happened

Hy4 flash is Tencent's first flash-tier language model, arriving less than two weeks after the flagship Hy4. Built on the Hunyuan architecture, it has 224 billion total parameters, including 64 billion Engram parameters — a conditional memory technique that stores frequently occurring text patterns — and activates 12 billion per forward pass.

Pricing is the story of this launch. List API prices are 0.5 yuan per million input tokens, 2 yuan for output and 0.02 yuan for cache hits — about $0.07, $0.281 and $0.003 in dollars. Against Hy4's list prices, those rates are about 12 times lower on input, 9 times on output and 15 times on cache hits. Through 12:00 on November 30, the TokenHub API is half the list price and use inside WorkBuddy is free.

The model is live on Tencent Cloud's TokenHub, WorkBuddy, Yuanbao and other Tencent products, under the API id hy4-flash, with OpenAI- and Anthropic-compatible protocols. Tencent said it plans to release the model's weights and list it on OpenRouter, with no date given.

Key facts

  1. Parameters: 224 billion total parameters, including 64 billion Engram parameters; 12 billion activated per token
  2. Context: Multimodal input with a one-million-token context; up to 960,000 input tokens and 64,000 output tokens
  3. API pricing: List prices: 0.5 yuan per million input tokens, 2 yuan for output, 0.02 yuan for cache hits
  4. Task cost: On Tencent's internal WorkIndex benchmark, per-task cost is 1/20th of Hy4 and 1/8th of DeepSeek's V4.1-Flash-1017
  5. Satisfaction: In the same blind test, user satisfaction was 2.79 out of 4, close to Claude Haiku 5.5's 2.81
  6. Discounts and open weights: TokenHub API is half price and WorkBuddy free through November 30; weight release and an OpenRouter listing are planned, with no date

Why it matters

Hy4 flash pushes Hunyuan's price down to 0.5 yuan per million input tokens, adding a low-cost tier two weeks after the flagship; the cost figures come from Tencent's internal benchmark and are not independently verified.
Useful Tap if this story helped you
Send to a friendMakes a share image for X, WeChat, Telegram and more

SourcesTechNode (Nov 16, 2026); VectorWire (Nov 16, 2026). This article is compiled from public information and does not constitute investment advice.

From HIMIAO Horizon · himiao.ai

Grab todayOpenAI Codex for Students: $100 in Free CreditsHIMIAO Picks →

Follow HIMIAO Horizon

No sign-up · Unsubscribe any time

Email

Notifications

A ping on your phone or computer when the brief is out. No email needed.

HIMIAO PicksWeekly freebies: one email every Friday, a 5-minute read

Comments

  1. Loading comments…

Related

  1. Nov 5Tencent Hy4 vs Mistral Large 4: Open-Weight Flagships a Week Apart, One's Weights Out and the Other's Pending, Output Prices $0.401 Apart
  2. Nov 4Tencent Releases Full Hy4 With Vision and Computer-Use Capabilities
  3. Nov 5Oracle Cancels $7B Tencent Chip Lease as Pressure From Both Sides Ends Remote-Compute Deal
  4. Oct 22DeepSeek Raises About $15 Billion, With CATL and Tencent Among Biggest Investors
  5. Oct 21ByteDance, Alibaba and Tencent Place Over $35 Billion in Nvidia Chip Orders, Buying Up China-Compliant RTX PRO 5500

Today

Nov 17 Tuesday

Latest news

All →
  1. Nov 17OpenAI Launches Science Blog as GPT-6 Astra Leads Proof of Candidate Lattice Standard Model Definition
  2. Nov 17Apple Smart Glasses Said to Debut at WWDC 2027 With No Display, Leaning on Siri and iPhone
  3. Nov 17AI Labs Race to Hire Counterintelligence Staff: Anthropic 33 This Year, OpenAI 22
  4. Nov 17Trump Backs Huang's Return to Nvidia: 'He Learned a Big Lesson'
  5. Nov 16Kuo: OpenAI and Anthropic to Buy About 26M Google TPUs in 2028–29
  6. Nov 16Huawei Cloud Launches Ascend-Native Embodied World Model CloudRobo-WM1.2
Ask the cat