·HSK 4

China
Zhōngguó
中国
artificial intelligence
réngōng zhìnéng
人工智能
company
gōngsī
公司
DeepSeek (Chinese AI company)
DeepSeek
DeepSeek
to release / to publish
fābù
发布
brand new
quánxīn
全新
version 4
V4
V4
flagship
qíjiàn
旗舰
AI model
móxíng
模型
。

DeepSeek (Chinese AI company)

DeepSeek

at / on (time preposition)

yú

于

April 24, 2026

2026 nián 4 yuè 24 rì

2026年4月24日

to release

fābù

发布

(completed action particle)

了

version 4

series

xìliè

系列

preview version

yùlǎn bǎn

预览版

，

including

bāokuò

包括

V4 Flash (lightweight model)

V4 Flash

and

hé

和

V4 Pro (flagship model)

V4 Pro

two models

liǎng kuǎn

两款

open source

kāiyuán

开源

model

móxíng

模型

。

V4 Pro (flagship model)

V4 Pro

to possess / to have

yōngyǒu

拥有

1.6 trillion

yīdiǎn liù wànyì

一点六万亿

(measure word)

gè

个

parameters

cānshù

参数

，

shì

是

currently

mùqián

目前

the largest

zuìdà

最大

(descriptive particle)

的

open source

kāiyuán

开源

model weights

quánzhòng

权重

model

móxíng

模型

。

April 25, 2026 at 4:00 AM

China

Zhōngguó

中国

artificial intelligence

réngōng zhìnéng

人工智能

startup company

chūchuàng gōngsī

初创公司

DeepSeek (Chinese AI company)

DeepSeek

at / on (time preposition)

yú

于

April 24, 2026

2026 nián 4 yuè 24 rì

2026年4月24日

to release

fābù

发布

(completed action particle)

了

version 4

flagship

qíjiàn

旗舰

series

xìliè

系列

(descriptive particle)

的

preview version

yùlǎn bǎnběn

预览版本

。

this time / on this occasion

cǐ cì

此次

release / launch

fābù

发布

to include

bāokuò

包括

two models

liǎng kuǎn

两款

model

móxíng

模型

：

V4 Flash (lightweight model)

V4 Flash

and

hé

和

V4 Pro (flagship model)

V4 Pro

，

both / all

jūn

均

to use / to adopt

cǎiyòng

采用

open source

kāiyuán

开源

approach / method

fāngshì

方式

to release

fābù

发布

。

V4 Pro (flagship model)

V4 Pro

to possess / to have

yōngyǒu

拥有

1.6 trillion

yīdiǎn liù wànyì

一点六万亿

(measure word)

gè

个

total

zǒng

总

parameters

cānshù

参数

，

among which

qízhōng

其中

49 billion

sìbǎi jiǔshí yì

四百九十亿

(measure word)

gè

个

are / constitute

wéi

为

active

jīhuó

激活

parameters

cānshù

参数

，

shì

是

currently

mùqián

目前

the largest

zuìdà

最大

(descriptive particle)

的

open source

kāiyuán

开源

model weights

quánzhòng

权重

model

móxíng

模型

。

in addition / furthermore

cǐwài

此外

，

DeepSeek (Chinese AI company)

DeepSeek

to launch / to introduce

tuīchū

推出

(completed action particle)

了

called / named

míng wéi

名为

Hybrid Attention Architecture

hùnhé zhùyìlì jiàgòu

混合注意力架构

(descriptive particle)

的

new technology

xīn jìshù

新技术

，

can / is able to

kě

可

to cause / to allow

shǐ

使

model

móxíng

模型

in / during

zài

在

long conversations

cháng duìhuà

长对话

within / amid

zhōng

中

even more

gèng

更

accurately

zhǔnquè

准确

(adverbial particle)

地

to remember / to retain in memory

jìyì

记忆

content

nèiróng

内容

。

both models

liǎng kuǎn

两款

model

móxíng

模型

both / all

jūn

均

to support

zhīchí

支持

one million

yībǎi wàn

一百万

(measure word)

gè

个

token (in AI)

lìngpái

令牌

(descriptive particle)

的

context window

shàngxiàwén

上下文

length

chángdù

长度

，

among

zài

在

open source

kāiyuán

开源

model

móxíng

模型

among / within

zhōng

中

to be classified as

shǔyú

属于

a first / pioneering achievement

shǒuchuàng

首创

。

regarding / in terms of

zài

在

pricing

dìngjià

定价

aspect

fāngmiàn

方面

，

V4 Flash (lightweight model)

V4 Flash

per / each

měi

每

one million

bǎiwàn

百万

input

shūrù

输入

token (in AI)

lìngpái

令牌

to charge a fee

shōufèi

收费

$0.14 US dollars

0.14 měiyuán

0.14美元

，

V4 Pro (flagship model)

V4 Pro

whereas / in contrast

zé

则

is / amounts to

wéi

为

$1.74 US dollars

1.74 měiyuán

1.74美元

。

both models

liǎng kuǎn

两款

model

móxíng

模型

already

yǐ

已

to upload

shàngchuán

上传

to / onto

zhì

至

Hugging Face (AI model hosting platform)

Hugging Face

platform

píngtái

平台

，

developer

kāifāzhě

开发者

can / may

kě

可

for free

miǎnfèi

免费

to download

xiàzài

下载

model weights

quánzhòng

权重

and / also

bìng

并

on / in

zài

在

locally / on local machine

běndì

本地

to run / to operate

yùnxíng

运行

。

New Words

DeepSeek

DeepSeek (Chinese AI company)

qíjiàn

旗舰

flagship

cānshù

参数

parameters

jīhuó

激活

active

kāiyuán

开源

open source

quánzhòng

权重

model weights

hùnhé zhùyìlì jiàgòu

混合注意力架构

Hybrid Attention Architecture

cháng duìhuà

长对话

long conversation

zhǔnquè

准确

accurately

dìngjià

定价

pricing

News in English

Science & Technology

DeepSeek Unveils New V4 Flagship AI Model

DeepSeek on April 24, 2026 released a preview of its V4 series, including the open-source V4 Flash and V4 Pro models. V4 Pro has 1.6 trillion total parameters, making it the largest open-weight model available.

Chinese AI startup DeepSeek released a preview version of its flagship V4 series on April 24, 2026.

This release includes two models: V4 Flash and V4 Pro, both using open-source distribution methods.

V4 Pro possesses a total of 1.6 trillion parameters, with 49 billion constituting active parameters, representing the largest open-source model weights currently available.

Additionally, DeepSeek introduced a new technology called Hybrid Attention Architecture, enabling the model to remember content more accurately during long conversations.

Both models support a context window of one million tokens, which is pioneering among open-source models.

Both models are now available on Hugging Face, where developers can freely download weights and run locally.

In terms of pricing, V4 Flash charges $0.14 per million input tokens, while V4 Pro is $1.74.

China
Zhōngguó
中国
artificial intelligence
réngōng zhìnéng
人工智能
company
gōngsī
公司
DeepSeek (Chinese AI company)
DeepSeek
DeepSeek
to release / to publish
fābù
发布
brand new
quánxīn
全新
version 4
V4
V4
flagship
qíjiàn
旗舰
AI model
móxíng
模型
。

New Words

News in English

DeepSeek Unveils New V4 Flagship AI Model

More in Science & Technology

United States
měiguó
美国
Congress
guóhuì
国会
approves
pīzhǔn
批准
artificial intelligence
réngōng zhìnéng
人工智能
regulation
jiānguǎn
监管
framework
kuàngjià
框架
。

OpenAI (AI company)
OpenAI
OpenAI
to launch
tuīchū
推出
GPT-6 (AI model)
GPT-6
GPT-6
and
yǔ
与
intelligent
zhìnéng
智能
interface
jièmiàn
界面
feature
gōngnéng
功能
。

Microsoft
Wēiruǎn
微软
to launch / to release
tuīchū
推出
Work IQ (Microsoft enterprise AI tool)
Work IQ
Work IQ
，
to let / to allow
ràng
让
Copilot (Microsoft AI assistant)
Copilot
Copilot
to connect to
liánjiē
连接
enterprise data
qǐyè shùjù
企业数据
。

ChinaZhōngguó中国artificial intelligenceréngōng zhìnéng人工智能companygōngsī公司DeepSeek (Chinese AI company)DeepSeekDeepSeekto release / to publishfābù发布brand newquánxīn全新version 4V4V4flagshipqíjiàn旗舰AI modelmóxíng模型。

New Words

News in English

DeepSeek Unveils New V4 Flagship AI Model

More in Science & Technology

United Statesměiguó美国Congressguóhuì国会approvespīzhǔn批准artificial intelligenceréngōng zhìnéng人工智能regulationjiānguǎn监管frameworkkuàngjià框架。

OpenAI (AI company)OpenAIOpenAIto launchtuīchū推出GPT-6 (AI model)GPT-6GPT-6andyǔ与intelligentzhìnéng智能interfacejièmiàn界面featuregōngnéng功能。

MicrosoftWēiruǎn微软to launch / to releasetuīchū推出Work IQ (Microsoft enterprise AI tool)Work IQWork IQ，to let / to allowràng让Copilot (Microsoft AI assistant)CopilotCopilotto connect toliánjiē连接enterprise dataqǐyè shùjù企业数据。

China
Zhōngguó
中国
artificial intelligence
réngōng zhìnéng
人工智能
company
gōngsī
公司
DeepSeek (Chinese AI company)
DeepSeek
DeepSeek
to release / to publish
fābù
发布
brand new
quánxīn
全新
version 4
V4
V4
flagship
qíjiàn
旗舰
AI model
móxíng
模型
。

United States
měiguó
美国
Congress
guóhuì
国会
approves
pīzhǔn
批准
artificial intelligence
réngōng zhìnéng
人工智能
regulation
jiānguǎn
监管
framework
kuàngjià
框架
。

OpenAI (AI company)
OpenAI
OpenAI
to launch
tuīchū
推出
GPT-6 (AI model)
GPT-6
GPT-6
and
yǔ
与
intelligent
zhìnéng
智能
interface
jièmiàn
界面
feature
gōngnéng
功能
。

Microsoft
Wēiruǎn
微软
to launch / to release
tuīchū
推出
Work IQ (Microsoft enterprise AI tool)
Work IQ
Work IQ
，
to let / to allow
ràng
让
Copilot (Microsoft AI assistant)
Copilot
Copilot
to connect to
liánjiē
连接
enterprise data
qǐyè shùjù
企业数据
。