OpenAI (AI company)OpenAIOpenAIto releasefābù发布GPT-5.4 (OpenAI language model)GPT-5.4GPT-5.4，to supportzhīchí支持one millionbǎiwàn百万tokenscí yuán词元context windowshàngxià wén上下文andbìng并cankě可to operate / controlcāokòng操控computerdiànnǎo电脑。
OpenAI (AI company)
OpenAI
OpenAI
to release
fābù
发布
GPT-5.4 (OpenAI language model)
GPT-5.4
GPT-5.4
to support
zhīchí
支持
one million
bǎiwàn
百万
tokens
cí yuán
词元
context window
shàngxià wén
上下文
and
bìng
并
can
kě
可
to operate / control
cāokòng
操控
computer
diànnǎo
电脑
OpenAI (AI company)
OpenAI
OpenAI
on (formal)
yú
于
March 5, 2026
2026 nián 3 yuè 5 rì
2026年3月5日
released
fābù
发布
completion marker
le
了
latest
zuìxīn
最新
flagship
qíjiàn
旗舰
model
móxíng
模型
GPT-5.4 (OpenAI language model)
GPT-5.4
GPT-5.4
the new model
xīn móxíng
新模型
supports
zhīchí
支持
up to
gāodá
高达
one million
yī bǎi wàn gè
一百万个
tokens
cí yuán
词元
structural particle
de
的
context window
shàngxià wén chuāngkǒu
上下文窗口
and
bìng
并
can
kě
可
autonomously
zìzhǔ
自主
to operate
cāokòng
操控
computer
diànnǎo
电脑
and
hé
和
web browser
liúlǎn qì
浏览器
March 5, 2026 at 8:00 PM
OpenAI (AI company)
OpenAI
OpenAI
on (formal)
yú
于
March 5, 2026
2026 nián 3 yuè 5 rì
2026年3月5日
officially
zhèngshì
正式
launched
tuīchū
推出
completion marker
le
了
its
qí
其
latest
zuìxīn
最新
flagship
qíjiàn
旗舰
model
móxíng
模型
GPT-5.4 (OpenAI language model)
GPT-5.4
GPT-5.4
the (formal)
gāi
该
model
móxíng
模型
passive marker
bèi
被
is positioned as
dìngwèi wéi
定位为
currently
mùqián
目前
the most powerful
zuì qiángdà
最强大
the most efficient
zuì gāoxiào
最高效
structural particle
de
的
professional work
zhuānyè gōngzuò
专业工作
frontier model
qiányán móxíng
前沿模型
in
zài
在
API (application programming interface)
API
API
version
bǎnběn
版本
in
zhōng
中
GPT-5.4 (OpenAI language model)
GPT-5.4
GPT-5.4
supports
zhīchí
支持
up to
gāodá
高达
one million
yī bǎi wàn gè
一百万个
tokens
cí yuán
词元
structural particle
de
的
context window
shàngxià wén chuāngkǒu
上下文窗口
setting
chuàng
创
a historic new high
lìshǐ xīn gāo
历史新高
compared with
yǔ
与
the previous generation model
qián dài móxíng
前代模型
compared
xiāngbǐ
相比
the new model
xīn móxíng
新模型
in
zài
在
individual claims
dān tiáo xìnxī
单条信息
error rate
cuòwù lǜ
错误率
in terms of
shàng
上
reduced by
jiàngdī le
降低了
33 percent
33%
33%
overall
zhěngtǐ
整体
responses
huídá
回答
containing errors
chūcuò
出错
structural particle
de
的
probability
gàilǜ
概率
also
yě
也
decreased by
jiǎnshǎo le
减少了
18 percent
18%
18%
in addition
cǐwài
此外
GPT-5.4 (OpenAI language model)
GPT-5.4
GPT-5.4
integrated
zhěnghé le
整合了
code generation
dàimǎ shēngchéng
代码生成
advanced reasoning
gāojí tuīlǐ
高级推理
and
yǔ
与
desktop control
zhuōmiàn cāokòng
桌面操控
and other
děng
等
multiple
duō xiàng
多项
capabilities
nénglì
能力
the model
gāi móxíng
该模型
can
kěyǐ
可以
autonomously
zìzhǔ
自主
browse web pages
liúlǎn wǎngyè
浏览网页
operate
cāozuò
操作
desktop applications
zhuōmiàn chéngxù
桌面程序
this capability
zhè yī gōngnéng
这一功能
is called
bèi chēngwéi
被称为
computer use
jìsuànjī shǐyòng
计算机使用
ability
nénglì
能力
OpenAI (AI company)
OpenAI
OpenAI
at the same time
tóngshí
同时
also released
tuīchū le
推出了
smaller
gèng xiǎo de
更小的
GPT-5.4 Mini (smaller model)
GPT-5.4 Mini
GPT-5.4 Mini
and
hé
和
Nano (smallest model tier)
Nano
Nano
versions
bǎnběn
版本
to meet
yǐ mǎnzú
以满足
different
bùtóng
不同
users
yònghù
用户
structural particle
de
的
needs
xūqiú
需求
analysts
fēnxī rénshì
分析人士
point out
zhǐchū
指出
this time
cǐ cì
此次
release
fābù
发布
further
jìnyībù
进一步
expanded
kuòdà le
扩大了
OpenAI (AI company)
OpenAI
OpenAI
in
zài
在
AI (artificial intelligence)
AI
AI
field
lǐngyù
领域
structural particle
de
的
competitive advantage
jìngzhēng yōushì
竞争优势
News in English
Science & TechnologyOpenAI Releases GPT-5.4 With One-Million-Token Context Window and Native Computer Use
OpenAI released GPT-5.4 on March 5, 2026, its most capable frontier model to date. The model supports context windows of up to one million tokens and can autonomously operate computers and web browsers.
More in Science & Technology
MicrosoftWēiruǎn微软and / withyǔ与OpenAI (AI company)OpenAIOpenAIto revise / to amendxiūɡǎi修改cooperation / partnershiphézuò合作agreement / termsxiéyì协议，to give up / to relinquishfànɡqì放弃exclusivedújiā独家license / authorizationshòuquán授权。
Microsoft
Wēiruǎn
微软
and / with
yǔ
与
OpenAI (AI company)
OpenAI
OpenAI
to revise / to amend
xiūɡǎi
修改
cooperation / partnership
hézuò
合作
agreement / terms
xiéyì
协议
to give up / to relinquish
fànɡqì
放弃
exclusive
dújiā
独家
license / authorization
shòuquán
授权
Microsoft and OpenAI Revise Partnership to End Exclusive Licensing Arrangement
Dell (company)Dài ěr戴尔first quarterdì yī jìdù第一季度AI (artificial intelligence)AIAIserverfúwùqì服务器revenueyínɡshōu营收year over yeartónɡbǐ同比to surge dramaticallybàozēnɡ暴增757 percent757%757%。
Dell (company)
Dài ěr
戴尔
first quarter
dì yī jìdù
第一季度
AI (artificial intelligence)
AI
AI
server
fúwùqì
服务器
revenue
yínɡshōu
营收
year over year
tónɡbǐ
同比
to surge dramatically
bàozēnɡ
暴增
757 percent
757%
757%
Dell AI Server Revenue Surges 757 Percent Year Over Year in Record Quarter
United KingdomYīngguó英国researchersyánjiū rényuán研究人员to obtain / to gain access tohuòdé获得GoogleGǔgē谷歌Willow (Google quantum chip)WillowWillowquantumliàngzǐ量子chipxīnpiàn芯片access rights / right to useshǐyòngquán使用权。
United Kingdom
Yīngguó
英国
researchers
yánjiū rényuán
研究人员
to obtain / to gain access to
huòdé
获得
Google
Gǔgē
谷歌
Willow (Google quantum chip)
Willow
Willow
quantum
liàngzǐ
量子
chip
xīnpiàn
芯片
access rights / right to use
shǐyòngquán
使用权
UK Researchers at Kings College London Gain Access to Google Willow Quantum Chip
United StatesMěiguó美国Illinois (US state)Yīlìnuòyī Zhōu伊利诺伊州to pass (legislation)tōngɡuò通过across the US / nationwidequánměi全美most stringent / strictestzuì yángé最严格AI (artificial intelligence)AIAIsafety / securityānquán安全bill / legislationfǎàn法案。
United States
Měiguó
美国
Illinois (US state)
Yīlìnuòyī Zhōu
伊利诺伊州
to pass (legislation)
tōngɡuò
通过
across the US / nationwide
quánměi
全美
most stringent / strictest
zuì yángé
最严格
AI (artificial intelligence)
AI
AI
safety / security
ānquán
安全
bill / legislation
fǎàn
法案
Illinois Passes Nation Most Stringent AI Safety Bill Requiring Third-Party Audits