汉字News

← 汉字NewsChinese News for Learners
technology
kē jì
科技
·HSK 5

DeepSeek (Chinese AI company)
DeepSeek
DeepSeek
to release / to publish
fābù
发布
open-source
kāiyuán
开源
large AI model
dà móxíng
大模型
V4 (version 4)
V4
V4
，
performance / capability
xìngnéng
性能
to rival / to match
pìměi
媲美
Western
xīfāng
西方
top-tier / leading
dǐngjí
顶级
model
móxíng
模型
。

DeepSeek (Chinese AI company)
DeepSeek
DeepSeek
on (a date)
yú
于
April 24th
sì yuè èrshísì rì
四月二十四日
to release
fābù
发布
open-source
kāiyuán
开源
flagship
qíjiàn
旗舰
model
móxíng
模型
V4 (version 4)
V4
V4
，
to include / to contain
bāohán
包含
two
liǎng gè
两个
versions
bǎnběn
版本
。
the said
gāi
该
model
móxíng
模型
on
zài
在
agentic / AI agent
zhìnéngtǐ
智能体
benchmark tests
jīzhǔn cèshì
基准测试
in / among
zhōng
中
structural particle
de
的
scores
défēn
得分
with / and
yǔ
与
GPT-5.5 (OpenAI model)
GPT-5.5
GPT-5.5
comparable / equivalent
xiāngdāng
相当
，
but / however
dàn
但
inference
tuīlǐ
推理
cost
chéngběn
成本
far lower than
yuǎn dīyú
远低于
the latter
hòuzhě
后者
。

May 7, 2026 at 8:00 AM

DeepSeek (Chinese AI company)
DeepSeek
DeepSeek
on (a date)
yú
于
April 24th
sì yuè èrshísì rì
四月二十四日
officially / formally
zhèngshì
正式
to release
fābù
发布
flagship
qíjiàn
旗舰
large AI model
dà móxíng
大模型
V4 (version 4)
V4
V4
，
and / also
bìng
并
simultaneously
tóngbù
同步
open-sourced
kāiyuán
开源
。
this time / on this occasion
cǐcì
此次
release
fābù
发布
includes
bāohán
包含
two
liǎng gè
两个
variants
biàntǐ
变体
：
parameter count
cānshùliàng
参数量
is / amounts to
wèi
为
1.6 trillion
1.6 wànyì
1.6万亿
structural particle
de
的
Pro (version)
Pro
Pro
edition / version
bǎn
版
，
as well as
yǐjí
以及
the lighter-weight
jiào qīngliàngjí de
较轻量级的
Flash (version)
Flash
Flash
edition
bǎn
版
。
both
liǎngzhě
两者
all / both
jūn
均
to support
zhīchí
支持
up to / as high as
gāodá
高达
one million
100 wàn
100万
(measure word)
gè
个
tokens
cíyuán
词元
structural particle
de
的
context window
shàngxiàwén
上下文
length
chángdù
长度
。
in / on
zài
在
agentic
zhìnéngtǐ
智能体
benchmark tests
jīzhǔn cèshì
基准测试
in / among
zhōng
中
，
DeepSeek V4 Pro (AI model)
DeepSeek V4 Pro
DeepSeek V4 Pro
structural particle
de
的
scores
défēn
得分
with
yǔ
与
GPT-5.5 (OpenAI model)
GPT-5.5
GPT-5.5
and
jí
及
Claude Opus 4.7 (Anthropic model)
Claude Opus 4.7
Claude Opus 4.7
comparable / on par
xiāngdāng
相当
。
however / yet
rán'ér
然而
，
its
qí
其
inference
tuīlǐ
推理
cost
chéngběn
成本
is only / amounts to just
jǐn wèi
仅为
competitors
jìngzhēng duìshǒu
竞争对手
structural particle
de
的
a small fraction
yī xiǎo bùfen
一小部分
。
as of / up to
jiézhì
截至
end of May
wǔ yuè dǐ
五月底
before
qián
前
，
promotional
cùxiāo
促销
pricing
dìngjià
定价
makes / causes
shǐ
使
Pro (version)
Pro
Pro
edition
bǎn
版
compared to
bǐ
比
post-promotion
cùxiāo hòu
促销后
structural particle
de
的
standard price
biāozhǔn jiàgé
标准价格
lower
dī
低
five times
wǔ bèi
五倍
。
analysts / commentators
fēnxī rénshì
分析人士
believe / consider
rènwéi
认为
，
this move
cǐjǔ
此举
further / additionally
jìn yībù
进一步
to intensify / to exacerbate
jiājù
加剧
completed action particle
le
了
Chinese and Western
zhōng xīfāng
中西方
AI (artificial intelligence)
AI
AI
companies
gōngsī
公司
between
zhījiān
之间
structural particle
de
的
competition
jìngzhēng
竞争
。

New Words

dà móxíng
大模型
large AI model
kāiyuán
开源
open-sourced
biàntǐ
变体
variants
cānshùliàng
参数量
parameter count
cíyuán
词元
tokens
shàngxiàwén
上下文
context window
tuīlǐ chéngběn
推理成本
inference cost
jìngzhēng
竞争
competition
jiājù
加剧
to intensify
cùxiāo
促销
promotional

News in English

Science & Technology

DeepSeek Releases Open-Weight V4 Model Rivaling Western Frontier AI

DeepSeek released its open-source flagship model V4 on April 24, offering two variants with over one trillion parameters and performance comparable to leading Western models at a fraction of the inference cost.

DeepSeek officially released its flagship large AI model V4 on April 24th and simultaneously open-sourced it. This release includes two variants: the Pro version with 1.6 trillion parameters, and the lighter-weight Flash version. Both support a context window length of up to one million tokens.

In agentic benchmark tests, DeepSeek V4 Pro achieves scores comparable to GPT-5.5 and Claude Opus 4.7. However, its inference cost is only a small fraction of competitors.

As of late May, promotional pricing makes the Pro version five times cheaper than the standard post-promotional price. Industry analysts believe this move further intensifies competition between Chinese and Western AI companies.

More in Science & Technology

Honor
róngyao
荣耀
launches
tuīchū
推出
robot
jīqìrén
机器人
smartphone
shǒujī
手机
equipped with
pèizhì
配置
motorized
diàndòng
电动
gimbal
yúntái
云台
camera
shèxiàngtou
摄像头
。

Honor Launches Robot Phone with Motorized Gimbal Camera Technology

domestically produced
guóchǎn
国产
chip
xīnpiàn
芯片
CXMT memory chipmaker
CXMT
CXMT
surge
biāoshēng
飙升
500 percent
wǔbǎi percent
500%
become
chéng
成
domestic
guónèi
国内
market value
shìzhí
市值
highest
zuìgāo
最高
stock
gǔpiào
股票
。

CXMT Memory Chip Company Surges 500 Percent, Becomes Most Valuable Stock on Mainland Chinese Exchanges

United States / American
Měiguó
美国
space agency / NASA
yǔháng jú
宇航局
to find / to discover
zhǎodào
找到
new
xīn
新
method / way
fāngfǎ
方法
to extend / to prolong
yáncháng
延长
Voyager 2 (spacecraft)
Lǚxíngzhě 2 hào
旅行者2号
to use / operation
shǐyòng
使用
lifespan / service life
shòumìng
寿命
。

NASA Engineers Find New Way to Extend Voyager 2 Mission After 49 Years in Space

global / worldwide
quánqiú
全球
semiconductor
bàndǎotǐ
半导体
second quarter / Q2
dì-èr jìdù
第二季度
sales revenue
xiāoshòué
销售额
to set (a record)
chuàng
创
all-time record high
lìshǐ xīngāo
历史新高
。

Global Semiconductor Sales Hit Record 403 Billion Dollars in Q2 2026